[2020-05-21] Version 0.8.13 - Internal changes: * Adjust for apt 1.9.11 pkgDPkgPM should not be used directly, use _system->CreatePM() instead; also compare map_pointer in cache against new MapPointer() method instead of Index(). (Patch by Julian Andres Klode; from the 0.8.12-2 upload; Closes: #953306) * Handle _error becoming a smart pointer thingy This should be fixed nicer in apt, so that we can use _error again somehow, as _GetErrorObj() is not really part of the API (but the ABI). (Patch by Julian Andres Klode; from the 0.8.12-3 upload) * Switch from libparse-debianchangelog-perl to libdpkg-perl The former was merged into the latter some time ago, with the main difference being the lack of some of the output formats that are not being used by aptitude. This seems to have been attempted in the past, but was reverted due to performance problems. The performance from current libdpkg-perl should be comparable. We use the perl module directly via a new aptitude-changelog-parser wrapper to avoid pulling dpkg-dev and its dependencies. This new wrapper will also make sure the module is available before proceeding. (Patch by Guillem Jover; Closes: #934135) * Rename README to README.DOCUMENTATION. - Translation updates: * po/nl.po: Dutch program translation by Frans Spiesschaert (Closes: #909687, #950541, #955506) * aptitude-defaults.ru: Russian sections translation by Sergey Alyoshin (Closes: #914786) * doc/po4a/po/it.po: Italian documentation translation by Beatrice Torracca (Closes: #959895) * doc/po4a/po/fr.po: French documentation translation by Jean-Pierre Giraud (Closes: #942094) [2019-08-25] Version 0.8.12 - New features: * Permit to retrieve changelogs of packages from Debian Ports repositories. * Adapt for changes in apt 1.9, like addressing deprecation and API changes. Thanks to Julian Andres Klode. - Bug fixes: * Adapt code to adapt to change of behaviour in Boost.MultiIndex library, and so we can start using Boost >= 1.66, 1.67 is about to become the default in Debian unstable. Thanks to Julian Andres Klode. * Fix crashes when Label of a repository is empty (Closes: #907436) * Fix changelog downloads * Read the correct number of elements in pkg_item::pkg_columnizer::init_formatting() -- it was ignoring the last element, reading only 25 out of 26. - Internal changes: * Subsitute readdir_r() by readdir(), the former is deprecated. src/generic/util/dirent_safe.h was removed due to being unnecessary after this change. * Use APT::PrettyDep(cache, dep) to avoid warnings about deprecation of dumping dep directly to std::ostream * Avoid deprecation warning by adding second parameter in TimeRFC1123() invocation in download_queue.cc. Parameter set to "false" following code in apt's pkgAcqIndex::Custom600Headers(). * Fix for missing "#include " (necessary for upcoming change in cwidget) [2018-07-29] Version 0.8.11 - New features: * Add search term for release label (?label) (Closes: #892983) by Tomas Pospisek - Documentation: * Change en/README_encoding to use utf-8 as all the other languages, instead of iso-8859-1 - Internal changes: * Use MKDIR_P instead of MKINSTALLDIRS in the building system, the latter was lost when m4/gettext.m4 was removed, and it was deprecated anyway. * Pass "-no-home" to elinks when creating text files from HTML, to avoid the annoying errors when creating the localised documentation (with "terminal beep" included): "ERROR at home.c:156: Unable to find or create ELinks config directory. Please check if you have $HOME variable set correctly and if you have write permission to your home directory." * Use elinks only if available (similar to po4a, rsvg-convert, etc.) * Switch to several std:: facilities from C++17's instead of Boost - std::filesystem instead of boost::filesystem - std::optional instead of boost::optional * Fixes needed by GCC-8 and more strict versions of the library - Translation updates: * zh_CN.po: Simplified Chinese translation by Boyuan Yang (Closes: #885916) * ru.po: Russian translation by Lev Lamberov (Closes: #893027) * Russian documentation translation by Lev Lamberov (Closes: #893028) [2017-11-15] Version 0.8.10 - User visible changes: * [Guillem Jover] Update section descriptions: Add "golang". Remove long-obsolete "non-US". - Bug fixes: * Restore ability to download changelogs with recent libapt (Closes: #881422) There were changes in the backend of downloads of changelogs, so the URI can be store:// instead of gzip:// since (lib)apt 1.2, and has to be since in recent uploads. So it was just switched to use store:// unconditionally, backporting to Jessie at this point is not high in priority... - Internal changes: * Remove m4/ with embedded copies of building-system helpers. The debian part of the package will depend on autoconf-archive to provide those needed. [2017-08-19] Version 0.8.9 - Bug fixes: * Fix crash due to wrong string pointer being printed (Closes: #865346) * Fix compilation error with GCC 7 (Closes: #853316) In a conversion involving an "operator string()", it seems that the code gets confused with the available ABI of std::string, similarly to what had happened with the ABI change of GCC-5 (but this is not supposed to happen in this new release). The error was: .../generic/apt/download_update_manager.cc:133: undefined reference to `URI::operator std::__cxx11::basic_string, std::allocator >()' - Translation updates: * Dutch documentation translation by Frans Spiesschaert (Closes: #867980) [2017-06-11] Version 0.8.8 - User visible changes: * Avoid spurious warnings about unused code paths (Closes: #863755) - Documentation improvements: * Replace minimal example for configuring resolver hints with a full example. (Closes: #702565, part 1) * Document keywords "maximum" and "minimum" for cost levels. (Closes: #702565, part 2) * Mention in the man page that a tilde character in front of an order keyword reverses the order. (Closes: #814038) - Translation updates: * da.po: Danish translation by Morten Bo Johansen (Closes: #861087) * cs.po, aptitude-defaults.cs: Czech translation by Miroslav Kure (Closes: #861940) * nl.po: Dutch translation by Frans Spiesschaert (Closes: #862922) * Dutch documentation translation by Frans Spiesschaert (Closes: #861709) [2017-04-18] Version 0.8.7 - User visible changes: * Warn about invalid locales (Closes: #859907) - Bug fixes: * Preserve auto-installed flag with hold/unhold/keep operations (Closes: #843536) * [cmdline] Fix extreme slowness of keep-all (Closes: #842707) Every change of marking the package to "keep" was triggering a reevaluation in aptitudeDepCache::end_action_group() of what had changed ("mark and sweep", duplicate of cache, triggers of packages state changes... etc), for every package. * Avoid problems or improve response in problems related to reinstalling (Closes: #851901) Improvements to deal with several problems that sometimes appear when reinstalling or acting on scheduled reinstall actions, due to the versions disappearing from the repositories. * [cmdline] Failing to apply actions are not fatal with -f / Aptitude::CmdLine::Fix-Broken (Closes: #835372) The behaviour of refusing to continue when there are errors happens since the change "[cmdline] Abort with Failure when actions cannot be taken (Closes: #121313, #430392, #445034, #498239, #576212, #639789, #798320)", but it's not an optimal solution for automatic installers like FAI, so this is a way to revert to the old behaviour. - Internal changes: * new method is_version_available(), to check if the given version of a package is available (to check availability e.g. for reinstalls) - Translation updates: * Italian documentation translation by Beatrice Torracca (Closes: #858784) [2017-03-05] Version 0.8.6 - User visible changes: * [cmdline] Better message when there are no deb-src in sources.list (Closes: #841875) - Bug fixes: * Use a more strict mask for comparison of dependency operators, thanks to Aaron M. Ucko for the initial patch (Closes: #836567, #837946, #849370) Recently apt started to use more values in an enum used for dependency operator comparisons, which causes some problems because aptitude's code from many years ago, which did not expect new values. Using a stricter mask to capture only the operators, as per the documentation of the enum (lower 4 bits). * Fix reading tag and their classification in subtrees (Closes: #853037) Newline characters were added to the last tag (or the only one), and it caused extra problems like classifying the tag in their own subtree due to the new-line mismatch in the tag name. * Fix misdetection of auto-installed packages from apt (Closes: #831858, #841347) - Translation updates: * ru.po: Russian translation by Lev Lamberov (Closes: #855329) * Dutch documentation translation by Frans Spiesschaert (Closes: #854897) [2017-01-24] Version 0.8.5 - Translation updates: * ru.po: Russian translation by Ulyanich Michael and Lev Lamberov (Closes: #849761) * nl.po: Dutch translation by Frans Spiesschaert (Closes: #851888) * Dutch documentation translation by Frans Spiesschaert (Closes: #851890) * French documentation translation by Cédric Boutillier (Closes: #852352) [2016-12-27] Version 0.8.4 - User visible changes: * Updated section list and added descriptions. Thanks Josh Triplett! (Closes: #847530) * Support for Build-Depends/Conflicts-Arch, by David Kalnischkies (Closes: #843789) - Bug fixes: * [curses] Do not fail immediately and attempt to continue when some package lists cannot be updated (Closes: #834790) The bug, in which the update of package lists stopped and the interface was left at the "Loading cache" stage indefinitely, seems to be due to changes in behaviour of apt or some strange interaction, because the relevant sections of aptitude have not changed in a long time. This is fixed now by not being strict with this kind of errors (to cope with what seems to be a new behaviour from libapt -- to not report that the update can proceed)... Hopefully this change will not create other problems. - Internal changes: * Support recent versions of googletest and google-mock, change of paths: /usr/src/gtest -> /usr/src/googletest/googletest /usr/src/gmock -> /usr/src/googletest/googlmock - Translation updates: * cs.po: Czech translation by Miroslav Kure (Closes: #838790) * help-cs.txt: Czech translation by Miroslav Kure (Closes: #838790) * nl.po: Dutch translation by Frans Spiesschaert (Closes: #841116) * aptitude-defaults.ru: Russian translation of updated section list by Lev Lamberov (Closes: #847965) * Dutch documentation translation by Frans Spiesschaert (Closes: #842186) [2016-08-07] Version 0.8.3 - User visible changes: * [cmdline] Commands "markauto", "unmarkauto", "hold", "unhold", "keep", "keep-all" and "forbid-version" now work independently of other action commands and they will not try to go ahead with pre-scheduled actions (Closes: #142699, #567079, #764299) This is similar to how "forget-new" or user-tag related commands worked, and the fact that they started to install or remove packages caused confusion. - Bug fixes: * [curses] Use single-line for "forget-new" dialog when using minibuf prompts (Closes: #832363) * [cmdline] "download", "source" and others provided by a thin wrapper over apt tools now accept "-t" and other options (Closes: #832320) * [cmdline] Document "source" and "showsrc" in --help (Closes: #833584) * [cmdline] Saying 'n' to a solution jumps to the end of the generated solutions and tries to generate a new one (Closes: #832907) This reverts b41db12f ("Rejecting a solution advances to the next one only") but also improves the description. * Fix code preventing to save package selection states to dpkg (Closes: #825866, #833588) - Internal changes: * Create on_apt_errors_print_and_die(), for the common case of checking for errors from apt, printing them on the screen and exiting immediately with an error code * cmdline_util: Add get_packages_from_set_of_strings(), to help parsing package names and resolve patterns from the command line (using the lower-level get_packages_from_string()) [2016-07-23] Version 0.8.2 - New features: * [cmdline] New commands "showsrc" and "source" (Closes: #243412, #403372, #403631) Provided by a thin wrapper over apt tools. * [cmdline] "forget-new" accepts package names and pattern arguments Addresses part of #421043 (and merged). * [curses] "forget-new" accepts package names and pattern arguments (Closes: #421043) - Bug fixes: * [curses] Apply solution ('!') guarded against crash when there are no solutions (Closes: #823734) * [cmdline] Command "download" now provided by a thin wrapper over apt tools (Closes: #431568) This command from aptitude did not provide verification of downloaded files. * [curses] Flat view changed to honour ::Pkg-Display-Limit if exists (Closes: #814037) * Fix crash caused in infer_reason.cc for pkg_unused_remove (Closes: #824441) Introduced in some difficult to reproduce conditions by fix for "Enhance infer_reason() to explain more cases of pkg_unused_remove", bug #266061, in 0.8. * [cmdline] Fix for code that failed to fetch the changelog for source packages. Thanks Emmanouil Maroudas for the fix. (Closes: #827133) - User visible changes: * [curses] Change TUI message after "pressing 'q' and then 'Enter'" to no more mention "shutting down". Use "quitting" instead. - Internal changes: * aptcache.{h,cc}: Split aptitudeDepCache::forget_new() in two parts, one to unmark as 'new' a given set of PkgIterators, another unmarking all. * cmdline_util: Add get_packages_from_string(), to help parsing package names and resolve patterns in the command line - Documentation: * Guide/User's Manual: * Enhance the documentation of the use of "architecture specification" for ?architecture search pattern arguments, instead of simple string matches (Closes: #823927) The commit d78de1ea in 0.6.8.2-1 had the side effect that the strings for the arch search pattern are no longer treated as simple "string patterns" as it happened before, when architectures were matched with partial strings (as it does with package names, etc.). In the case of TUI with interactive search enabled, it means that packages do not match until the arch-specification is complete/valid, triggering "beeps" meaning that no packaged match. * Add clarification for ?any-version search pattern (Closes: #691531) * manpage: * Remove outdated exceptions of -t/--target-release for "changelog", "download" and "show" (Closes: #693684, #821866) * Describe aptitude-[create|run]-state-bundle as for internal use only (Closes: #521190) - Translation updates: * Russian documentation translation by Lev Lamberov (Closes: #829681) [2016-05-05] Version 0.8.1 - Bug fixes: * Install Essential or Required packages with the commands "full-upgrade" (command line) and MarkUpgradable ('U' in the curses interface) (Closes: #555896, #757028) * Do not prevent to remove/purge aptitude if already removed or purged (Closes: #822331) E.g. aptitude:i386 in amd64 with multiarch. * [curses] Indent section descriptions in subtrees (Closes: #103416) Not the same as creating a column, but same effect. * Invoke the resolver for unfulfilled Recommends only once (Closes: #822329) After the fix for #819636, the resolver was invoked after every action taken (in curses mode) to see if there were any unfulfilled dependencies. This is too slow for some people/systems, so this is now only invoked will full check of dependencies (Recommends / "soft deps") before the final action. * [curses] Be more verbose when quitting after dpkg actions (Closes: #822560) This is related to the fix in the previous version for #246672. Make curses windows visible when restoring (instead of staying blank) to show message of "Updating state and shutting down", and make some improvements to speed up shut down. The optimisation is achieved through a global variable shutdown_in_progress that apt_load_cache() pays attention to, to e.g. avoid re-loading tags. This mechanism is not very elegant, but: a) we cannot quit directly (we need to save the state after package actions were taken, for example), and b) untangling the rest of the shutdown process is very difficult, with the state of the code having grown organically for years without taking this into account (it would need significant rewrites); and duplicating code for the cases when we quit and we continue is not very good either - Translation updates: * ja.po: Japanese translation by Takuma Yamada (Closes: #822208) * nl.po: Dutch translation by Frans Spiesschaert (Closes: #822889) * Dutch documentation translation by Frans Spiesschaert (Closes: #822890) [2016-04-20] Version 0.8 - New features: * Detect and suggest solutions when pkgstates is corrupt (Closes: #405506) * Allow to choose between localized logs or not (default is not) with the option ::Localized-Log (Closes: #357828, #596221) * [curses] Left/Right arrow keys collapse/expand one level of the subtree, and when in a package item (row) Left jumps to "parent" (Closes: #157984, #241945, #415449) The solution is a bit "hacky", because for example for the left arrow in the package item to jump to its parent, the pkg_item objects have to capture events of some keybinding ("collapse") and re-inject the event as "parent". The direct route, which would be to add "left" as an alias of "Parent", doesn't work in this case, because some elements using "left" for other purposes would be confused and would interpret it as "Parent". Maybe there is another way to achieve this more cleanly if it's possible to assign bindings like "Parent" specific to e.g. tree items (pkg_items derive from that), but even if cwidget allows for this, it seems to be quite complicated to achieve it. * [curses] Be able to Quit after dpkg run (Closes: #246672) The current implementation seems to work, but the handling of these issues is a bit brittle and with lots of circular dependencies, with signals and events happening everywhere. E.g view (cwidget objects) being reloaded automatically when cache is reloaded, while cwidget is supended for dpkg run. The initial events trigger further updates and triggers using cwidget objects and other structures, sometimes depending on different options being enabled or how the curses mode was launched (e.g. "visual preview", or "solution screen"). In any case, the cache needs to be reloaded (and with it, state saved to disk) to perform some updates/cleanup after dpkg runs, like resetting "reinstall" state when done (the "save pending reinstall" was implemented in this version), or unmarking upgraded packages as needing upgrade after the version has been upgraded to the desired one (bug fixed in this version, when aptitude was not acknowledging the "upgrade" as having taken place, and still marking it as update in the next runs). Presumably all of these complications are why this hand't been implemented before, in all the intervening years. Still I thought that it was worth to implement it now because it will now at least save the user/system from spending time in some of the curses actions, save a few extra keystrokes, and with it a few seconds (specially in slower systems) having to keep an eye just for quitting. - Bug fixes: * ~E pattern now returns packages with Essential only (Closes: #548505) apt marks itsef as Essential, so this doesn't change much some aspects of the bug report. * Detect as "Security Upgrades" when site matches security.*.d.o (Closes: #328620) Previously it was only matching "security.debian.org", not accounting for mirrors like "security.eu.debian.org". * Allow to run in dumb terminals when not requiring curses preview or solution screen (Closes: #817276) The bug fixing #317928 forbade some uses which are not problematic. Also added a few more checks to also forbid launching solution screens and previews from the command-line. * "reinstall" planned action is now preserved across sessions, when becoming root or in the command line with --schedule-only (Closes: #255587, #785641) After the successful reinstallation of this package (and probably other installations/upgrades/deletions/... in the same action), the package is unmarked for reinstallation. There's no way to detect individually whether a package was reinstalled successfully, so if there is a failure in the whole set of actions, the package will still be marked for reinstallation even if the reinstallation itself has actually succeeded. This is the best that we can do without significant complications (e.g. handling reinstallations in a package-per-package basis and separate from other actions, but this would be somewhat similar to going back to the current situation -- not treating it just as other actions). It is unlikely that this is a big problem in the real world, though. Probably reinstallations are performed very frequently, and when they are, it's unlikely that they are bundled in conjunction with package actions such as big upgrades. * Fix crash when invoking curses resolver from command line (Closes: #817776) This fix is analog to the one for #816322, and previously to that the problems involving boost::flyweights. * Update internal state for upgrades without target version after the action is performed (Closes: #721426) This extends "Update internal state for upgrades/downgrades with target version after the action is performed (Closes: #787658, #714429)", by also trying to remove the "Upgrade: yes" line in pkgstates when the package has been upgraded to the candidate version (when a explicit version is not requested). Otherwise the lines linger in "pkgstates" for longer than necessary, until it is updated due to other packages needing changes; with the side effect that if an "aptitude update" happens before the lines are removed, the packages are automatically marked for upgrade in subsequent aptitude sessions. * [curses] Fix for miscalculated Download Size field (Closes: #817547) After running dpkg run or under some other conditions (cache closed), the object storing fetch information did not get reset and updated correctly. * Marking as keep, through ':' or other actions, clears Forbidden version Uncovered investigating #615481. * Fix cases where user tags operations did not apply (Closes: #819002) This was caused by misuse of locks before the operation of saving the state file (pkgstates). * Actions involving unfulfilled Recommends invoke the resolver (Closes: #819636) With APT::Install-Recommends="true" (the default), actions resulting in unavailable Recommends (e.g., the recommended package is upgraded instead of remaining in the required version) invoke the resolver, for the user to take decision. This is because now the resolver monitors situation when the PolicyBrokenCount is not zero, before it was only taken BrokenCount into account (it was the only count available when the code was set up). * Enhance infer_reason() to explain more cases of pkg_unused_remove (Closes: #266061) Enhance infer_reason() to explain more cases where package is removed because of automatically installed and unused (one case related to #819636). The case related with #819636 is when the version of the depended package (the one for which we look the reason of removal) changes because e.g. it is upgraded, so maybe it doesn't fulfill the original dependency (for example because the package depending on it it depends on an exact or less-than version, as in the bug report). The second case (#266061) was missing the explanation because it didn't take into account reverse-depends of virtual packages provided by the package being inspected. - User visible changes: * [cmdline] Output in dumb terminals, when there is no piping/redirection involved and even if ioctl doesn't work to get window size (e.g. inside emacs), will now use COLUMNS instead of the default very big default width * Format string scapes for Source (%E->%e) and Architecture (%e->%E) swapped This is for consistency while addressing #743769, while it's not too late (only a few releases in unstable since they were added). * New short form "~e" for search term "?source-package" (Closes: #743769) * [curses] Remove option to run "dpkg-reconfigure" on a given package (Closes: #680334, #686626, #738350) Similar to "reportbug" recently removed, it was mostly an annoyance with several related bugs open, with others fixed in the recent past (#474876), was not integrated with the rest of actions -- no menu entry, would trigger an immediate without confirmation (sometimes by mistake) instead of a planned action, since it was immediate could not be saved as most/all actions are... In addition, this is a relatively dangerous operation that should seldom be needed, and it's better if people run it run by hand rather than if they trigger it by mistake. Lastly, as a curiosity and as telling of the typical evolution of the sofware features, the history of it is: Added in the version 0.2.3 released on 2001-07-11, with this comment: "You can now run dpkg-reconfigure from within aptitude (it's a silly feature, but it's also about 10 lines of code..)". The original 10 lines of code are now 67 in pkg_item.cc, 1 keybinding in defaults.cc, mentions in the documentation and all of the translations; needs to use different mechanisms for "su-to-root" permissions that it needs, there were at least 4 bugs involving the feature, and spending a few hours tracking all of this to finally remove it... so much for an innocent silly feature with only 10 lines of code :-) * [cmdline] Solution screen for "Remove the following packages" shows version and suite This is to make it more similar to the other cases shown, which include a version, as well as being able to deduce when a package is local / obsolete ("now") instead of being part of other suites. E.g. it will show: Remove the following packages: 1) libsdl1.2-dbg [1.2.15+dfsg1-2 (now)] instead of the previous: Remove the following packages: 1) libsdl1.2-dbg - User visible changes (Resolver): This subsection is to separate the (many) changes of the Resolver in this version from other changes. When dealing with conflicts or when called when special options (--full-resolver in the command line), aptitude fires an internal resolver to try to address the resolution of conflicts caused by the requested action(s). This was created at a time when apt was not resolving very well the dist-upgrades and other actions, and long before Multi-Arch was implemented in Debian, so with time the gap between aptitude and others was smaller, often apt does a better job than aptitude, then unfavorable comparisons started to be made, and other problems started to appear in the implementation. This is a very complicated piece of software, and it's very difficult to fully understand and reason about the consequences of deep changes. As a heuristic system, the solutions that it gives are quite open (so it's better if it's guided towards desired solutions), and tweaks to the process can have unpredictable and undesired effects somewhere else. That said, since in the last few years there were many complaints about the first solution offered being very often undesirable (removal of many packages), and (from most user's point of view) preferable solutions like upgrades being offered only after many removals, and keeps only offered as last resort; some changes were made to try to address these problems. The main change that seems to have started this in the past was with #565867 in 0.6.1.5 back in 2010 (but there are bugs open much older than that, so it's been a recurring problem): + [all] Changed the default resolver configuration so that removals and "safe" actions are grouped together. Without this change, aptitude was being too conservative and failing to solve simple upgrades. (Closes: #565867) The safe-upgrade resolver is unaffected by this change, since it uses explicit constraints to prevent removals from being installed. In the comment of the bug report, the original developer says: "In the 0.6 series, the aptitude resolver is a lot more conservative than it used to be, due to repeated complaints by users that it was removing too many packages in the first solution that was presented. It looks like maybe I overcompensated, though (this is not the first report along these lines). The main difficulty in cases like this is getting the algorithm to properly balance removals with upgrades. However, it looks like the old scoring system works just fine here." This is still true now. Hopefully after the changes in this version the balance is acceptable for most common actions when using aptitude, including day-to-day upgrades as well as occasional big upgrades between Debian releases. Now onto the changes. Executive summary: * Changes to the order in which solutions are offered in conflict resolutions, so upgrades or keeps are preferred over removal of many packages (Closes: #341963, #359171, #365644, #453935, #569315, #570377, #574132, #588202, #610845, #651410, #651947, #653284, #661678, #722211, #798240) A more detailed description of individual changes follows: * Modify several resolver scores config variables ::ProblemResolver::RequiredScore was lower (4) than ImportantScore (5), even if it's in fact higher in the rank; while ExtraScore was -1, even if many regular package in current systems are priority "extra" -- they should not be penalised as "could we do without". The new scores are: ::ProblemResolver::RequiredScore = 8 (was: 4) ::ProblemResolver::ImportantScore = 4 (was: 5) ::ProblemResolver::StandardScore = 2 (was: 3) ::ProblemResolver::OptionalScore = 1 ::ProblemResolver::ExtraScore = 0 (was: -1) * Add new score for removal of obsolete package and corresponding config variable ::ProblemResolver::RemoveObsoleteScore Removal of obsolete packages is treated as positive (310) to allow for easier upgrades. It has to counter at least the penalty to remove packages (currently -300). * Add new score for partial solutions cancelling the removal of a package and corresponding config variable ::ProblemResolver::CancelRemovalScore Resolver actions cancelling the removal (or purge) of a package requested to be removed/purged were not being penalised, now they are (default -300). This was missing compared to penalisations for other actions like RemoveScore, when removing packages not marked for removal. Now, solutions of the resolver cancelling explicitly requested removals are also penalised. * Only add Preserve{Auto,Manual}Score if the packages are actually installed Without this, the algorithm was assigning extra score +60 (default PreserveManualScore) for packages which remained uninstalled (were not to be installed) and when the user actually didn't select them manually -- even if the justification was "because it is the to-be-installed version of a manually installed package". This caused for example that in amd64 with multi-arch 386 "Score: +60 for uno-libs3:i386 [UNINST] because it is the to-be-installed version of a manually installed package (..::PreserveManualScore)", and "Score: -20 for uno-libs3:i386 5.1.1-1 because it is a new install (..::InstallScore)", so installing the version rather than keep it uninstalled had comparatively a -80 disadvantage, 4 times worse than the penalty of installing the new package itself (which is already quite big). At the same time, since PreserveAutoScore is 0, the package which was actually installed had "Score: +0 for uno-libs3:amd64 5.1.1-1 because it is the to-be-installed version of an automatically installed package (..::PreserveAutoScore)". So this is now only applied to installed packages, to avoid these undesired effects. * Change the default score of several ::ProblemResolver config variables These are: ::StepScore to -10 (was +70) ::PreserveManualScore to +20 (was +60) ::UpgradeScore to +30 (was 0) Long explanations: 1) ::StepScore to -10 (was +70) Due to #418385 (back in 2007) it was raised from +10 to +70. There were already complaints after the change about "removing way too much stuff on upgrades". With positive numbers it gives extra score for exploring more and more steps in the problem space (more inclined to search "depth-first" in the graph), and the comparatively large score meant that this was offsetting the negative score of some generally undesired actions (e.g. removing packages which are otherwise fine), so the overall balance of taking more actions on packages was positive even if it meant removing many packages removed rather than upgrading only one. (On a related note, RemoveObsoleteScore was added to consider removal of local/obsolete packages as positive, when e.g. doing dist-upgrades and leaving packages behind). This played a big role in the problem (that appeared or became worse at some point a few years ago) of the first suggestions offered by the full resolver to be those removing many packages, rather than upgrading a single one or a few, or at least suggesting the one(s) causing the conflict (selected to be upgraded by the user) to be kept. This has been tested in several bread-and-butter scenarios with a better outcome than the previous solutions (dist-upgrade, and upgrading packages that breaks others if not upgrading them --transitions, dependencies on same versions of the same source package, etc--). It will of course have some side effects, and perhaps has to be reverted or further tweaked, but at least is an attempt to improve the general situation. 2) ::PreserveManualScore to +20 (was +60) As explained in another point of this version in the changelog, this variable caused that the score for installing a new version was -20 plus +60 bonus to keep the "manual" uninstalled version uninstalled. This was changed to only have into account cases where packages are installed, but it affects still other cases in similar ways. It also interacts badly with UpgradeScore, because if the score assigned to this variable is higher than the one in ::UpgradeScore, the package is preferred to be kept in the current versions than upgraded, and this used to cause big penalties for situations packages that need to be upgraded to the same versions. Upgrading packages coming from the same source in lock-step is a typical case where packages have these strict dependencies with exact versions. So it seems sensible to lower this score to something that might be more easily overcome by other package actions that are needed very often. 3) ::UpgradeScore to +30 (was 0) Apart from the explanation in 2) (that this score should be higher than ::PreserveManualScore to have better suggestions in many common cases), it seems natural that upgrades are preferred to actions at 0 like Keep, since very often the purpose of launching package management tools like aptitude is to upgrade -- if there are upgrades available. * Change the default value of ::ProblemResolver::Keep-All-Level config variable ::Keep-All-Level to 10000 (was 20000) This cost level is assigned to solutions that cancel all the user's actions ("keep all"). A cost level that is lower is preferable in the "safety" mode, which is the default. More information about Costs in the User's Manual documentation, section "Resolving package dependencies". There are two levels which were lower (at 10000 both) than this one, Safe-Level and Remove-Level. The first is for safe actions, the other one (also with 10000 in the last few years) is presumably to allow for packages to be removed when they become obsolete, chiefly in dist-upgrade situations. Since several years ago, in cases where the full resolver had to kick in, the first suggestion for conflicts or unfulfilled dependencies was often to remove a lot of packages. Part of the reason why this was preferred to upgrades or installations of new packages had to do with other causes (see other changes in the same version of the changelog). But the reason why all of those were preferred to the "keep packages uninstalled" (again, in the default settings with "safety" mode) is because this class of solution is ranked in the ::Keep-All-Level, which is higher than the others and thus left as the last resort. It was assumed that "keep packages uninstalled" was undesirable because of cancelling all of the user's requested actions of packages. At the same time, the solution where "the package that you requested should be removed rather than upgraded, and along this one this dozen too" was at ::Remove-Level, in consequence, always offered before ::Keep-All-Level. It seems natural to have "keeps" being offered at the same "safety" level than removal of many packages, and in fact it should be offered before removing many packages, including ones that are installed and requested to upgrade and trigger the conflict that is being resolved. This change makes that almost all day-to-day solutions are classified at the same level now, excluding exceptional cases like removing essential packages or installing non-default versions. When at the same level, solutions are classified by an extra score depending on the actions taken, which is a cause of further tweaks in this version, so different solutions at the same safety level are not offered randomly in any case. But perhaps this change doesn't leave the "safety" levels mechanism in a very good position, if almost all solutions are always in a single level. In any case, I think that it's better to fix the oddity described above rather than overhaul a very complex system, and potentially causing many more unintended consequences than the one that this change might cause. * Change how the costs are compared Costs defer comparisons to "cost implementations" (cost_impl), and these were compared based on "structural_level", and when it was equal, on "actions". "actions" is a vector of pairs with "level indices" and "levels". For "levels" only the value is calculated. But before reaching that point, vectors and pairs were compared in a more generic way, so if the vector had a smaller number of elements and was equal up to that point, the shorter one was chosen as smaller (similar to lexicographical comparisons of strings), ignoring the effect of some modifiers in "actions". This was extremely convoluted to follow and reason about, so a more explicit comparison calculation is beneficial. Additionally, the overall result with the previous comparison is that it still produced undesired solutions in many cases, due to the levels of the solutions being the same ("safe", "remove", the newly changed "keep-all") when the resulting package changes (simple upgrades, removal of packages requested to be upgraded, keep current state) are perceived as very different in importance / desirability by the users. The modifiers in the "actions" change the level values a bit, but these were not taken into account in many cases in the previous comparison to give preference to some solutions over the others. Maybe these modifiers are not accurate or don't produce good results in all cases (they are untested for a number of years due to the behavior of comparisons until now, if nothing else), but the new sorting (with the combination of the other changes in this release, modifying scores and levels) seems to produce much better results overall. * Apply InstallScore and UpgradeScore also when the packages are not "manual" For some reason this was only enabled for "manual" packages, even if there was nothing in the documentation mentioning this. This is another step to promote Upgrades over other solutions. - Internal changes: * configure.ac: Allow to disable check for Boost library headers individually Note: Should be used only when developing, if at all. The reason to implement this is that the check is expensive, 1~2 minutes at ./configure time (more than 50% of the time that it takes to run the whole script); and if it's known that the packages and versions of related Boost libraries are there and working then it is unnecessary to check again and again. * Avoid extra code calling apt_load_cache() when already loaded src/cmdline/cmdline_util.cc src/ui_download_manager.cc The cache just (re)loaded in last steps of download_manager (either download_install_manager or download_update_manager). Loading again is mostly harmless because the function when called detects that it has already been loaded and quits, but duplicated code is not good in general, and the call was modified several times lately, so it is always a maintenance burden (not knowing / noticing that it was actually duplicated code, otherwise would have removed it then). * [cmdline] Rejecting a solution advances to the next one only Previously it was advancing to the last generated solution, which is equivalent at the moment (generates one by one), but not if/when several solutions are generated at once (which might happen in future changes). * [General] Update FSF's postal address everywhere, drop duplicated boilerplates. - Documentation: * manpage: * Minor clarification about "install" example (Closes: #268698) - Translation updates: * da.po: Danish translation by Morten Bo Johansen (Closes: #817065) * ja.po: Japanese translation by Takuma Yamada (Closes: #819603) * Actually enable Dutch (nl) documentation translation by Frans Spiesschaert [2016-03-05] Version 0.7.8 - Bug fixes: * Allow update to happen at the start ("-u") (Closes: #816385) While trying to make the code to behave the same in download_install_manager and download_update_manager, a change was made to make the latter to save the "package state" before proceeding. However, when aptitude is launched with "-u", download_update_manager runs before apt_cache_file is loaded for the first time, so it's not possible to save the "package state". So revert the behaviour and act as before -- try to save the state, but don't enforce it. * Extra check in update_pkgAcquire_fetch_info() to make sure that pkgRecords / apt_packagerecords is available * Now explains packages being removed when deps are broken (Closes: #342835) infer_reason() was not checking for this reason for package_auto_removed. * Avoid crash due to divide by zero in progress counters (Closes: #816431) * load_tags_from_verfiles(): guard against progress pointer being null * Use factory from APT to support different progress bars for dpkg actions (Closes: #816520) * Fix problem with recent change that prevented to work with local repositories (Closes: #816537) Even if we would like to not have to get locks and have to call fetcher->Run() if no (remote) fetch is needed (see #766122), it is needed anyway to work with local repositories (see #816537). * Fixing crash when invoking curses resolver from command line (Closes: #816322) The solution had already been found to deal with crashing problems involving boost::flyweights in the past (copied from "struct close_cache_on_exit" in main.cc). Instead of calling exit() directly, the solution is to invoke a sequence of shutting down data structures from apt and aptitude via apt_shutdown(). * Fix problem with auto-flag due to persistent Auto-New-Install in pkgstates (Closes: #816497) Auto-New-Install remained in pkgstates after the package was installed. As a consequence, (un)markauto (and maybe some other parts) were not working on packages where this had been recorded, because the persistent Auto-New-Install was read back upon start and was interfering with the logic of the code handling auto-flags. * Convert std::sprintf to std::snprintf (Closes: #767533) [2016-03-01] Version 0.7.7 - New features: * Only lock package cache dirs and call pkgAcquire->Run() when necessary (Closes: #766122) * [curses] Bindings for reject/approval of solutions in resolver can be applied to subtrees (Closes:#475595) * Several optimizations that improve performance at startup and when reading/writing files (Closes: #312920) See "Internal Changes" for more details. Basically, when using some basic command that shouldn't perform much computation ("aptitude install does-not-exist"), the time was cut down from around ~3s to ~1.5s in my system. The system is a relatively modern computer with SSD disks (but nothing bleeding edge), with multi-arch enabled (2 architectures) and close to 90 thousand binary packages, which is the double of systems without multi-arch enabled and affects many operations significantly (like doubling the size of the file "pkgstates"). * [curses] Download Size field shows remaining to be fetched (Closes: #446085) Now it will show "DL: 368 kB/661 kB" instead of "DL Size: 368 kB" when part of the files to download are already there, and the field is now aligned to the right. To make room for this in narrow terminals the field "Will use 43.0 kB of disk space" had been changed first to "Disk usage: +43.0 kB" due to other needs, and now to "Disk: +43.0 kB". The implementation was a bit tricky, though. The total size of downloads is pre-calculated in libapt and readily available; but to check for the actual size needed one needs to instantiate and initialize pkgAcquire, pkgSourceList and pkgDPkgPM, and then call a method to check for the files that are already there on the disk. The whole operation took only fractions of a second (~0.03s in the development system, in the testing conditions), but could be worse in other cases or when the disk is slow or busy. In command-line mode this only happen once, so the overhead is negligible; but on interactive systems this is bad because: a) it happens multiple times; and b) because interactivity and screen freezes --even if small-- do not mix well, and the program seems sluggish. But specially, due to how aptitude implements redrawing the screen, this was done in almost every key stroke, e.g. when moving to the next package/line down, so the overhead was very significant. So a mechanism was implemented (in previous commits) to only retrieve this information when the state of any package changes (marked to install, to delete, etc.), and to cache the result. Afterwards, on typical redraws of the screen without changes of packages involved (e.g. just scrolling), only the cached information is needed, and filling this field takes only 30 microseconds (in the development system) instead of ~30ms when full check is needed. - Bug fixes: * [cmdline] Fix crash with "show" and not installed packages (Closes: #815581) * [curses] Fix for wrong width/properties of some columns in the columnizer, incl. "hostname" rendered with 4 characters instead of 8 (Closes: #815554) Some columns like shortpriority, pin_priority and trust_state should not be present in default_widths due to their size being fixed and independent of translations. The mechanism is quite brittle, but maybe it's difficult to get something more robust since it involves getting width values that depend on width of translated strings. * [curses] Fix crash when trying to view changelog. Thanks Wei-Cheng Pan for the fix. (Closes: #815635) * [cmdline] In "search" and "versions", when width is set explicitly, pipes/redirections do not act as --disable-columns (Closes: #815690) Width can be set through config or command line. The behaviour to act as --disable-columns changed in the last version, 0.7.5; before it was using 80-width columns by default, or explicitly set width. * [curses] "%r" (revdepcount) now includes virtual packages (Closes: #320089) This was caused by the patch to address #270057, PkgIterator::ProvidesList() doesn't include virtual packages provided by the visible version (should use VerIterator::ProvidesList()). * [curses] infer_reason() for pkg_unused_removed now has recommends/suggests into account in all cases (Closes: #476596) It was not taking into account several cases when packages were being removed (because of being marked as auto-install and not required anymore), e.g. when the dependency was downgraded to Suggests, or only showing Recommends if config option APT::Install-Recommends was enabled. Even if APT::Install-Recommends is not enabled, if a package is installed it is kept as long as it's recommended (the user has decided to install it, despite being only recommended); and similar for Suggests; so the reason for removal should include these cases as explanation for being removed at this point. * Fixed case of packages wrongly being marked with auto-flag upon start (Closes: #816229) This was due to variable of extended-state previously_auto_package not being cleared when marking the package as manually installed, combined with using this variable to decide whether to write "Auto-New-Install: yes" in pkgstates in a recent change of 0.7.6, commit 7369d8f9 to address/fix #563877, which is another problem caused by the handling of automatic flags. * Packages to "keep" when applying solutions of the resolver maintain current manual/auto (Closes: #815551) In previous versions, first it was removing the auto flag unconditionally (#508428), probably due to APT thinking that it was a decision to keep it manually, and aptitude not marking it explicitly as auto. The fix for #508428 in recent versions changed this by setting the flag, but aptitude's resolver considers "keep" as a more generic "install_version", with new version being the same as current, and the auto-flag being always true in those cases. This is now refined to treat specially the "install" solutions when a package is actually to be kept -- first it retrieves the auto-flag of the current state of the package, then marks to keep, then sets the auto-flag accordingly after asking APT to keep the current version. - User visible changes: * [curses] Change "Will use/free %sB of disk space" for "Disk: %sB" Addresses some concerns of #815554, closed elsewhere. The new string is shorter (helps when terminal width is 80 columnsor so) and should be equally clear. * [curses] In Preview screen, bottom pannel, improve dependency reasons explanations for package to be installed (Closes: #814592) * [curses] Remove option to run "reportbug" on a given package or package version (Closes: #463510) It was mostly an annoyance for the reasons explained in the bug report: aptitude often running as root and reportbug complaining, often triggered by mistake, and it is not a very useful feature in general. * [cmdline] Operations like "show", "search", "versions", "why"/"why-not" or "changelog" do not need to lock dpkg state directories (Closes: #498061) Maybe some of the ones which still require it don't need it, but they are not used very often and leaving locking enabled is the default, safe option. - Internal changes: * Support for most changes in recent versions of apt (Closes: #687678, #687686, #687688, #687684) aptitude had to be adapted already in the last few months for many of the changes in recent versions of apt (up to 1.2.*), including some that prevented compilation or caused warnings/errors to be emitted when downloading changelogs, removed or deprecated API calls, and caused crashes in some of the most unfortunate cases. Keeping the above bugs open, several without any useful information, did not help to make the needed changes happen after 3.5 years around; and Wheezy and Jessie stable releases went out without these bugs being acted upon. Some of the important remaining problems or bugs with actual useful information will remain open. Several, like aptitude not handling config options '-o' correctly, are not blocking or related to "catch-up with changes in apt" at all, but problems in aptitude in its own right. In general, changes of apt will keep coming continuously, maybe making some of the changes in previous releases irrelevant, so as time goes on these older and generic "to-dos" without useful information make less and less sense to keep around. So marking those bugs as closed in this release -- the changes that they mention and that are not covered by specific bugs is basically complete, up to 1.2.*. * Move the definition methods of dpkg_selections to the headers (optimization) These methods are called inside loops many times (based on the number of packages, which are many tens of thousands), so it makes a difference in common operations which happen very often (e.g. more than 0.1s, when saving the file state). * Lazy initialization of reading "tasks" information (optimization) This functionality was called upon start from src/generic/apt/apt.cc, and taking more than 0.3s (in modern computer with SSD disks), when many times in command line mode is not needed. Now the loading is delayed until it's needed (when other related methods are called and expect this information to be present). * Optimizations when reading/writing files These operations happen typically several times every time that the program runs. The changes save 0.2 to 0.3s (in modern computer with SSD disks), with multi-arch enabled (2 architectures) and close to 90 thousand binary packages, in a simple test where an inexisting package is requested to install in the command line ("aptitude install does-not-exist"). The individual optimizations are: - Not updating progress for every package when reading/writing files, only ~10 times (every 10% increment) instead of ~90 thousand for the test scenario above - Declare some variables, specially strings, out of the loop executing for every package -- in that way, initialization and reserving memory is saved - Writing aptitude state file only when complete * Optimizations when reading debtags This functionality was called upon start from src/generic/apt/apt.cc. There are two modes, with "debtags" package installed and from "verfiles". Loading with "debtags" was improved from 0.13s to 0.05s, and "verfiles" from 0.4s to 0.27s, in modern computer with SSD disks, multi-arch with 2 architectures enabled. The improvement consisted in doing some extra calculations to update progress only ~10 times (every 10% increment) instead of ~54 thousand ("debtags") or ~121 thousand ("verfiles") for the test scenario above. * Optimizations when building trees for package views The tests were done in modern computer with SSD disks, multi-arch with 2 architectures enabled (close to 90 thousand packages to show). This functionality is used in when building the views in curses mode. Creating the tree for the initial default view takes ~0.42s, creating the tree for a new default "Package view" ~0.22s, "Debtags browser" ~0.3s, "Sources view" ~0.28s, "Flat" ~0.18s, "Package view" with limit "~slibs" (section "libs") ~0.09s. With optimizations enabled, the initial default view takes ~0.34s, creating the tree for a new default "Package view" ~0.13s, "Debtags browser" ~0.21s, "Sources view" ~0.23s, "Flat" ~0.12s, "Package view" with limit "~slibs" (section "libs") ~0.08s. The improvement consisted in doing some extra calculations to update progress ("Building view") only ~10 times (every 10% increment) instead of ~90 thousand for the test scenario above (~10 thousand with the limit "~slibs"). * src/generic/apt/apt.*: Create get_pkgAcquire_fetch_info() to reuse in the future Update the information when the package installation states change, and cache the information, so it can be used for example in the interactive mode without the information being retrieved again and again when the screen is redrawn (simply moving a line up or down, for example). * Make Progress parameters to be a pointer in several functions in aptcache.*, so it can be passed as null if necessary/desirable - Translation updates: * Dutch documentation translation by Frans Spiesschaert (Closes: #815183) * Dutch translation by Frans Spiesschaert (Closes: #815185) [2016-02-21] Version 0.7.6 - New features: * [cmdline] "show -v" shows all hashes, not only MD5Sum * Improvements to "why"/"why-not" (Closes: #729349) It now gives more information about installed versions, manually or automatic installations and priority; instead of "Unable to find a reason to install/remove" when no dependencies require or conflict with a pacakge. This enhancement is not only for the command-line commands "why" and "why-not" -- it is also for one of the info screens in the bottom half in curses/TUI (cycling with 'i'), and typing 'w PKG' in command line prompts when requesting confirmation. * Add support to clean the packages downloaded after successful install or similar operations (config option ::Clean-After-Install) (Closes: #110685) * [curses] "Keep" now works in version nodes (Closes: #585182) * [curses] "Install" on virtual packages now installs single candidate, or emits an error if there are no candidates or more than one (Closes: #174820) * [curses] Dependency subtrees are now sorted in a pre-determined order (Closes: #808882) * Show Origin URI for packages being untrusted when asking for confirmation (Closes: #582064) It does not explain why they are untrusted (libapt does not seem to provide much information) but at least is a better hint than just the package name. * [curses] In package info screen, "Packages depending on [this]" are further split based on the virtual (provided) packages that they depend upon (Closes: #159584, #760610) This helps to avoid the reverse dependencies listed multiple times within the same subtree, as it was happening until now; but specially to vastly improve the use-case to find out which packages depend not on this package directly, but on virtual packages also provided by this package (so perhaps an alternative can be installed). * [cmdline] "search" and "versions" do not truncate columns with redirections/pipes (Closes: #445206, #496728) It will behave as passing "--disable-columns" without column limit (large enough number for all practical purposes). - Bug fixes: * Properly set a default for $XDG_CACHE_HOME if not set. (Closes: #807545) * [cmdline] "w" at prompt now actually works (Closes: #555014) * Refuse to remove aptitude within aptitude (Closes: #319782, #568548) * [cmdline] Address prompt deficiencies related with translation and input methods (Closes: #475802) * Abort with Failure (rather to continue with the installation) when packages fail to be fetched -- except if APT::Get::Fix-Missing is present and set to true. When this happens, the actions in the command line also return with non-zero status. (Closes: #121313, #302784) * [cmdline] Abort with Failure when actions cannot be taken (Closes: #121313, #430392, #445034, #498239, #576212, #639789, #798320) The Closes of #121313 overlap with other fixes, there are reports of each case merged. * [curses] Adjust widths and "expand"/"shrink"-ability of some fields so Broken count and Download size are shown fully even with 80-width terminals (Closes: #810221) * [cmdline] Changelog operations now exit with non-zero on errors (Closes: #675833) * Fix problem showing versioned provides of virtual packages in error message It was showing candidate version or version of the provider "real" package instead of the version of virtual package ("versioned provides" is not a very commonly used feature yet, but there were requests to support them better). * Fix problem showing the actual version of the package providing the virtual package in error message It was showing the candidate version of the provider, rather than the version actually providing the virtual package -- the candidate version, and in general different versions, do not necessarily provide the same virtual packages. * Avoid loss of automatic flags when packages are marked for removal but not removed immediately, and the state is saved in between (Closes: #686573, #718802, #787663) Thanks to David Kalnischkies for the investigation and suggestion to fix this. It started to happen in aptitude after 2010 due to a change that happened in apt to fix #572364 (a change of a default in a parameter in pkgDepCache::writeStateFile() to ignore packages marked as deleted). * Avoid loss of automatic flags when package actions are scheduled multiple times before taking place (Closes: #563877) aptitude has a mechanism to store information about packages-to-be-installed that are pulled in automatically, when apt doesn't save that information (#435079, commit 1193257d). When scheduling actions multiple times, it was not taking this extra information into account, so the new scheduled actions ruined the automatically-installed flag of packages previously scheduled. * [cmdline] Arguments for "versions" which are not patterns now treated as '?exact-name' instead of '?name' (Closes: #691874) * [curses] "Cancel pending actions" now reloads the cache (roughly equivalent to restarting the program), rather than marking all packages as "keep" plus ruining all auto-installed flags and holds (Closes: #537735, #576319) * Reinstate auto-installed flag when marking packages to keep in apt cache (Closes: #508428) There were situations like for example in conflicts/solution resolver, that after applying the solution "Keep at current version", sometimes packages had their auto-installed flag cleared. * Message explaining unability to fetch changelogs now suggests "update" operation (Closes: #456822) * Do not use ncurses ("visual mode") in dumb terminals (Closes: #317928) * [cmdline] Do not show "update" progress in dumb terminals (Closes: #757940) * [cmdline] Catch exceptions when stdin is not available in prompts (Closes: #620782) * Brief explanation when column definitions could not be parsed (Closes: #813319) * Do not download changelogs when they are already being downloaded (Closes: #810020) Until now, pending changelogs kept being queued (e.g. pressing 'C' repeatedly in the curses interface spawned many downloads), and in the latest versions of apt (>= 1.1) it caused aptitude to crash. In previous versions of aptitude/apt I did not crash, but it caused weird errors ("download queue being destroyed"). Maybe internal changes within apt cause now crashes when they were errors before (not necessarily apt's fault, maybe it's a misuse from aptitude). The errors already seen in previous versions seem to be caused by aptitude code, for example because all downloads with the same URI are deleted from some internal structures of aptitude. With this "fix" at least repeated downloads are not repeated (*), which is a nice feature in general, and seems to solve this problem. (*) Still, sometimes the changelog is downloaded and shown twice, maybe this is because of sync problems between background threads, or maybe because sometimes the download happens so fast that the first is already downloaded by the time that the second is queued (this fix only prevents to queue when "to-be-started" or "active-downloads", not when it was downloaded previously). More in general, the code related with downloading files/changelogs is incredibly complex and entangled, spans many files and dozens of classes nested again and using inheritences and overlapping code, and it uses a home-grown implementation of threads in cwidget copied or based on POSIX threads. After several multi-hour sessions trying to analyse the point of the crash and improve this situation, and failing to wrap my head around it, I think that it's probably better to implement a new way to download changelogs, which I believe that can be simplified significantly, and hopefully in this case other problems will be easier to fix, and we can start to use pkgAcqChangelog and avoid the warnings about not being able to change the user to "_apt". But this has to wait for another release. * Fix so libapt can drop privileges to _apt:nogroup when downloading from the network and use pkgAcqChangelog (Closes: #797527, #806595) When downloading changelogs, aptitude was using temp::dirs which are owned by root or the user launching aptitude, and libapt could not drop privileges, causing unwanted annoying warnings (#806595). Since aptitude can run as root and normal users, it is also not possible to let libapt create files and copy them later: when running as normal users, it is not possible to access those files to copy them. This was changed by letting libapt to download to an intermediate place created by aptitude, and copy the file afterwards. Another change is that aptitude uses the specific functionality in libapt for changelogs, pkgAcqChangelog, present since apt v1.1. The whole set of classes dealing with download queues is very complex and should probably be rewritten to adapt to real needs and adapt to changes in apt over the years. This is a quick way to fix this annoying item present already for months, without ripping apart the whole download subsystem. See also comment of the other entry closing #810020. * Improve support for trace dumps (Closes: #814408) Probably due to changes in config variable handling from apt during the years, this functionality (related with ::ProblemResolver::Trace-Directory and ::ProblemResolver::Trace-File) was copying too many unrelated files, including parts of the root fs and causing the temporary directory to fill-up. Improvements included to copy only the needed files, as well as trying to safe-guard about similar problems with apt config values in the future. Also, removed an extra block of code copying a directory twice by mistake. - User visible changes: * [cmdline] Prompt dialog shows lowercase instead of capitalised "yes"/"no" (Closes: #607444) * [cmdline] "search" now exits with non-zero on errors or empty results (Closes: #497299) * Remove the trailing period from some error messages (Closes: #803011) * [curses] New style for Download Progress in Status bar, with higher contrast (Closes: #803980) * "why" modes, group-by modes and log levels should be provided in English, translations not accepted anymore (Closes: #738345) * [curses] Pressing keybinding "Keep" (":") on package names or versions does not remove the auto-installed flag * [cmdline] Show error message when using --purge (Closes: #434502) * [cmdline] "show" selects by default the candidate version (as documented in man page) and shows extra information when the package is installed but can be upgraded, rather than showing "not installed" (Closes: #539978) * [cmdline] Give more prominence to Version field in "show" command by showing it as second * [cmdline] Improve message about packages similar to mistyped names (Closes: #812397) - Internal changes: * configure.ac: change order so aptitude CXXFLAGS, CPPFLAGS and LDFLAGS override dependencies' (e.g. to set new compiler standards) * Use pkgDepCache::GetCandidateVersion() instead of GetCandidateVer(), deprecated apt API * Create function to convert from priority to a string and reuse in several places * New apt function: is_auto_installed(PkgIterator) * Lazy initialisation and better encapsulation of metadata download cache (Closes: #807852) This avoids some problems in builders, like pbuilder (not all env vars are cleaned, or for some reason the home of the user invoking aptitude inside pbuilder does not exist), and when home or some specific directories are not writable for some reason. This "metadata download cache" is only necessary for screenshots (GTK+, possibly Qt?, disabled for a long time and probably removed soon), and to store successfully downloaded changelogs after the user explicitly requests that operation in command line or TUI/curses mode -- which does not happen in every session, and never at all like when autobuilders use aptitude as resolver. * Add common function in apt.{h,cc} to clean cache of downloaded package files * Convert from some deprecated methods of libapt to supported ones * Add the common function (class method) add_msg_to_display() in download_list.{h,cc} to reuse code in several places * Move pkg_subtree_with_order to "pkg_subtree.h" * New util function: is_dumb_terminal() * New util function: print_ncurses_dumb_terminal() * New util function: create_temporary_changelog_dir() This functionality is similar to temp::file and temp::dir (and these were used before), but apt v1.1 added a feature by which the process drops privileges to _apt:nogroup when downloading files from the network, to not have code running as root doing such tasks. As a consequence, temp::dirs are owned by root and libapt cannot drop privileges, causing unwanted annoying warnings (#806595). Since aptitude can run as root and normal users, it is not possible to let libapt create files and copy them later: when running as normal users, it is not possible to access those files to copy them. This function supports other code to get around these problems. - Documentation: * manpage: * Explain that "why"/"why-not" need a valid package name (Closes: #535635) * Improve documentation of "forbid-version" (Closes: #773023) * Clarify automatic flag treatment with "install" actions (Closes: #469219) * Fix typo in --autoclean-on-startup (Closes: #812864) * Guide/User's Manual: * Make images available in the french translation by copying them from the original document (Closes: #414150, #697736) This is obviously suboptimal because the images are not translated, but since 10 of the 11 were missing altogether, this is not making the problem worse. The remaining image of the copied ones was present in the french translation, but was missing a key part of the interface described in the surrounding text (it was still "wrong", and hopefully the new one is more understandable in this context, even if in a different language). The suboptimal solution is being taken at this point because the bugs have been reported almost 9 years ago, and almost 2 years ago I requested help from the Debian French l10n team, and there has not been any feedback so far. Perhaps having English images in the French translation (rather than missing the images altogether) will spike somebody into action, to send us a recent and localised capture to fix the problem properly ;-) * Add note about ?upgradable (Closes: #498440) To explain that it just means that there are upgrades available, not related at all with "safe-upgrade" or "full-upgrade" operations - Translation updates: * Italian documentation translation by Beatrice Torracca (Closes: #808985) * Portuguese translation by Miguel Figueiredo (Closes: #810651) * Danish translation by Morten Bo Johansen (Closes: #811400) [2015-12-08] Version 0.7.5 - New features: * [curses] Show Origin and URI in pkg info screen (Closes: #294040, #457794, #513622, #611515) * [cmdline] Print version of provider of virtual package in conflicts/broken (Closes: #802943) * [cmdline] Show packages that will not be upgraded with verbose>0 (Closes: #553577) * [cmdline] Notify of 'Download Only' mode before confirmation to continue (Closes: #446265, #804287) * [cmdline] When simulating do not prompt for removal of essential packages (Closes: #613049) * [cmdline] When simulating, add notice before confirmation to continue actions (Closes: #173277) * Use XDG_CACHE_HOME spec instead of ~/.aptitude/cache (Closes: #671780) * Print useful message when package lists are out of date (Closes: #758764) * [curses] Allow to hide bottom half in several screens (Closes: #453853) * [curses] Show hostname on header line (Closes: #398101) * Log now distinguishes between remove and purge (Closes: #705612) * [cmdline] Improve error reporting of add/remove-user-tags (Closes: #652421) * Reuse format string %T also for user-tags (Closes: #498442, #665824) * Add format strings for Source (%E), Architecture (%e) and Origin (%O) (Closes: #248561, #604982, #760608) * [curses] Provide more info about broken packages in solution screen (Closes: #642800) This screen will now show the extra "(action, version/s involved)": The following packages depend on cpp-5 and will be broken by its removal: * cpp (upgraded, 4:5.2.1-6 -> 4:5.2.1-8) depends on cpp-5 (>= 5.2.1-13~) * gcc-5 (held/unchanged, 5.2.1-26) depends on cpp-5 (= 5.2.1-26) * Don't use __DATE__ & __TIME__ when printing version, for reproducible builds - Bug fixes: * [cmdline] Fix problems in 'build-depends' causing malfunctions (Closes: #802532) * Fix problem in which alternative dependencies of uninstalled packages were not resolved and satisfied properly, causing "Unable to satisfy dependencies", when they were actually installable. Due to a mistake/typo, the possible packages were always compared against versions of one of the alternatives (the last one), instead of comparing the proper package-version of each alternative. * Messages about build-depends conflicts (unable to satisfy dependencies) now show version comparison operators. There was a mistake in the variable to print (dependency type rather than operator). * [curses] Use default string for Default-Grouping when config is empty (Closes: #405970) * [curses] Improve message when becoming root fails (Closes: #516854) * [cmdline] For "search", use candidate or, if does not exist, current version rather than a random version to show information about the package (Closes: #490593) * Check that the config files are valid, otherwise exit (Closes: #472701) Upon reading the configuration files it did not check if they were valid, and for "~/.aptitude/config" it was writing it back immediately. The writing stopped at the point of the reading failure (instead of just skipping the problematic parts), so part of the previous configuration file was lost. It is safer (and not too onerous) to ask the user to fix the configuration file before continuing, rather than stomping on valid configuration values. * Use new config_file.{h,cc} to improve config file management (Closes: #504152, #502617, #442937, #407284) * When becoming root with 'su', pass along "--login" (Closes: #190784) * Return an error if adding/removing tags without "root" (Closes: #725272) * Format string '%t' tells about locally installed packages (Closes: #349413) * Save state before package reconfiguration (to not be lost) (Closes: #474876) * [curses] Allow to view changelogs of packages in Backports (Closes: #755677) This was already possible in command line. Now code was merged in command line and curses to work in the same way. * Get the real pin priority in "versions" and other cases (Closes: #640731) * Use new user-tags implementation, more secure (Closes: #792601) * Upgrades can do downgrades when pinned high (>1000, see apt_preferences) (Closes: #344700, #348679) "install" did the correct thing, but the differet upgrade modes (command-line and TUI/curses) did not consider to downgrade versions pinned higher than 1000, as established in apt_preferences. apt also behaves in this way, so it improves compatibility. Note that downgrading is highly discouraged in general, though. - User visible changes: * [cmdline] Minor improvements to help (Closes: #497136) * [cmdline] Hide Show-Why reasons when the string is empty (Closes: #576584) * [cmdline] Slightly improve and reformat --help * [cmdline] Document add/remove-user-tag in --help - Internal changes: * Change APT::Changelogs::Server (src/apt_options.cc) from http://packages.debian.org/changelogs to http://metadata.ftp-master.debian.org/changelogs -- it is the default for a few years already. * Add function to get the URI of a package version * Add function to get the Origin of a package version * Add function to get the candidate version of a package * Use Boost::Filesystem for some file-related operations C++ Filesystem TS is derived from it not approved yet, and the support from compilers is not great. When support is good, the switch should be almost automatic. * Add generic/apt/config_file.{h,cc} to help manage config file * attach/detach_user_tag() now return results of the operation * Create function get_user_tags() for compartimentalisation and reuse * Changes to support apt-1.1/libapt-pkg5.0 * Improve user-tags implementation and move to own file, away from aptcache - Documentation: * manpage: * Separate two elements in --add-user-tag-to description with comma, they were displayed together * Guide/User's Manual: * Add keybindings for menu bar (Closes: #365298) Add an indication in the document that "Control+[space]" and "F10" are also valid. [2015-10-16] Version 0.7.4 - New features: * [curses] Show providers of virtual packages in "description" area (Closes: #103964) The description area (bottom half of the screen in the default view) was completely empty for virtual packages. Now it shows the list of "real" packages providing the virtual package, in a nicely formatted list. - Bug fixes: * Retrieve again long descriptions (Closes: #801460) When disabling the #ifdef of HAVE_DDTP, it was done the wrong way around and the old code (untested for many years) compiled fine and didn't cause errors, but didn't retrieve long descriptions anymore (it did for short ones). * Fix for circular dependencies in internal_mark_delete() (Closes: #801430) Under some circumstances, when following reverse dependencies of packages to be deleted to see if they are automatically installed and unused (so they can be pro-actively marked for deletion as well), the function calls itself recursively. In this case, it uses this version with an extra parameter to detect when packages were already visited, to avoid infinite loops in the case of circular dependencies (bug #801430). * Fix race condition replacing ~/.aptitude/config (Closes: #764046) - User visible changes: * [cmdline] In versioned provides, include '=' symbol (Closes: #767393) At the moment, there is only one symbol ('='), and it is hardcoded in dpkg and apt, so doing the same to avoid possible confusion among users, as suggested by David Kalnischkies from APT Team. * [cmdline] "show" now shows arch of packages in multi-arch systems Until now, it was not appending the arch in Depends and other places, so there was output like: Package: dput-ng ... Conflicts: dput, dput Replaces: dput, dput Provides: dput Now this shows: Package: dput-ng ... Conflicts: dput, dput:i386 Replaces: dput, dput:i386 Provides: dput * [curses] Use 'w' in action column for package downgrades (Closes: #801310) In the column that shows the action selected for the packages (between current installation state and automatically installed), e.g. in the main package view and the preview screen before installing of the curses interface, there was no character for downgrades. There was 'u' for upgrades, 'd' for removals, 'p' for purges, 'h' for holds, etc., but the action character for downgrades was missing, so the column was left empty -- as it was an specific background colour until recently. * [cmdline] Show feedback of some commands (hold, forbid-version, ...) with verbose>0 (Closes: #270033, #638841) * [cmdline] Require args for some commands: hold, keep... (Closes: #759520) - Internal changes: * In internal_mark_delete(), also check for auto-installability property of real packages providing virtual packages to stop them being marked for deletion. can_remove_autoinstalled() checks for this anyway, but the implementation could change in the future, and it shouldn't hurt to double check. * Convert some instances of boost::unordered_{set,map,multimap} to std:: * Convert some instances of boost::hash to std:: * Remove two uses of boost::lexical_cast, use std::to_string() instead * Create utility function remove_non_recursive() - Documentation: * manpage: * Remove wrong assertion that safe-upgrade does not accept --full-resolver (Closes: #560991) * Document missing &BD specifier when installing packages (Closes: #696568) * Guide/User's Manual: * Update renamed config option ::CmdLine::Show-Summary (Closes: #776789) The config option used in the code is ::CmdLine::Show-Summary, but in one place in the documentation it was named ::CmdLine::Why-Display-Mode (probably from an incomplete rename). * Correct some wrong examples of search patterns (Closes: #519456) [2015-10-07] Version 0.7.3 - Bug fixes: * [cmdline] Correctly calculate the indentation for multi-byte encoded strings in command line "show" (Closes: #360202) * [curses] Show intended action when downgrading (Closes: #516517) In package view/info screen, it was not handling this action and not showing anything in this case. * Update internal state for upgrades/downgrades with target version after the action is performed (Closes: #787658, #714429) Thanks to David Kalnischkies and Julian Andres Klode from APT team for their help, clues and patience. It took some serious effort to get to this fix, which in the end is quite modest in terms of code changes. Internal state was neither reset nor written to disk until other packages' states changed, so aptitude marked the same action on the packages as pending to be performed again and again (but the action had already been taken, so there was no observable effect in terms of installations). However, this meant that the packages that were downgraded to older versions when newer were still available (e.g., mixing stable and unstable), was not shown as "Upgradable" neither in the current sessions nor in subsequent sessions (under Upgradable subtree in interactive mode; or command line like "aptitude search ~U"), until the file pkgstates was written to disk because of changes in other packages. * [curses] Don't precede Tags field by newline in package info screen (Closes: #799933) * Correctly detect state of vers/pkg when is ConfigFiles (Closes: #698768) * [cmdline] Show "RECOMMENDED but will not be installed" on pipes and redirs (Closes: #647474) The change was requested in #452202 in 2007 to not show this message and was applied, but after that there are at least 3 separate requests (#647474, #587676 and #720074) asking to show the same messages independently if the output is an interactive terminal, a pipe or a redirection to a file, all citing the lack of this message. (Makes sense, because it is one of the most prominent messages that users see, and also because this way of hiding messages is almost never used elsewhere in the code). In principle, "quiet" shouldn't make a difference when using ttys or pipes/files, but it does because it's primarily tied to the "progress" messages appearing all around the command line code; so in main.cc quiet is set to "1" when the output is not a tty. I think that the best solution by now is to revert the behaviour and simply show the message whenever there are packages to show. The only harm is if people complain again that it's too verbose, but probably nobody is using "-q" in practice to avoid having this message printed. A proper implementation would involve to clean-up the mess of quiet/verbose and convert it to only one single variable, and act verbosely accordinly, and perhaps tie the condition of showing progress of operations on another variable/mechanism. * [curses] Improve the integration of Forbid upgrades (Closes: #439258) Now it will advance the selection to the next line after marking a package with 'F', if that option is enabled. * Remove unneeded automatically installed packages in the same action (Closes: #478116, #564545, #637257, #368037, #486454, #738517, #789803, #779999, #756507, #759764, #766702, #655483, #740009) There were some cases where manually installed packages were removed/purged, but dependencies that were needed only by these packages (marked as automatically installed, and unused after the removal of the main package) were not removed as part of the the same "action", due to interactions between aptitude (in find_not_orphaned()) and libapt. The packages were detected as "garbage" in subsequent runs of apt (offering to "autoremove" them), aptitude in the command line, or further actions within the interactive session of aptitude; but nevertheless this caused numerous complaints about aptitude not removing the "cruft" (sometimes assuming that because it was not removed immediately, it will not be removed later either). Now, as part of the removal of a package and unless the option "::Delete-Unused" is set to true (defaults to false); aptitude pro-actively marks for removal the dependencies of the package marked as automatic and not used by other packages; and find_not_orphaned() was modified to only have into account installed packages or to-be-installed packages to prevent removals. * Honour ::Purge-Unused when removing packages (Closes: #724034) * Be more informative when virtual packages are involved in broken deps (Closes: #316633, #798898, #799918) * Fix description of config option ::AutoClean-After-Update (Closes: #756937) - User visible changes: * Make boundaries of sessions in the log file more clear (Closes: #269275) * [curses] Avoid marking as auto/non-auto if already set (Closes: #265013) * [curses] Print message when starting to perform actions (Closes: #323371) * Use uppercase APT:: for apt config vars, for consistency (Closes: #519425) Both APT:: and Apt:: (the other common form that we were using) are valid, in fact they are case-agnostic. But for internal consistency and specially for consistency with APT documentation (apt.conf), use the uppercase form. This mostly affects documentation, but also some code and translations. It was substituted automatically in all translations. * [cmdline] Print versions of Provides, if they exist (Closes: #767393) Some aspects of the bug report refer to the curses TUI and it is still unclear how to deal with them, so it was cloned to another report; but the change implemented for this part could be useful, it is harmless, and it simplifies code (two branches differentiating for this reason now merge). - Internal changes: * temp.cc: simplify code and messages and reduce the use of boost libs * configure.ac: Remove checks for boost/scoped_array.hpp and boost/random.hpp, not used anymore -- a few seconds less per compilation * In file pkg_ver_item.cc, function pkg_ver_columnizer::setup_column(), the switch/case handling for longstate was missing the options added in 2007 for APT_HAS_TRIGGERS, present for stageflag case and both cases in pkg_columnizer::setup_column() of pkg_columnizer.cc. (In other words, it was missing in 1 of these 4 places where it is necessary). * Disable obsolete checks in configure.ac and preprocessor checks in code for APT_HAS_TRIGGERS, APT_HAS_HOMEPAGE and HAVE_DDTP. The support for these functionalities has been present in all involved tools since 2007-2008 or earlier, so it is safe to disable the checks and the dead code. * Add functions to check if packages are installed, automatically installed, if they are virtual, and others. They can be used later to make other code more consistent, clear and clean. * Remove sigc++ include from apt_config_widgets.h (fails to build with newer versions) - Documentation: * Guide/User's Manual: * Document Brown colour for downgraded packages in UI (feature implemented in the last version) - Translation updates: * Russian by Lev Lamberov (Closes: #769425) * Dutch by Frans Spiesschaert (Closes: #800516) [2015-09-17] Version 0.7.2 - New features: * 'Escape' key cancels input dialogs (search, limit, etc.) (Closes: #493322) * Extend "firstchar" grouping policy to optionally use source package names. With "firstchar(binary)" or the default "firstchar" it will use binary package names (same behaviour as before); while with "firstchar(source)" it will use source package names. This is useful for example when browsing packages coming from the same source package ("source" grouping policy) within large collections (examples: all installed packages, all upgradable packages, all packages from "main" section), to add an intermediate grouping step. In this way, one can for example emulate the hierarchy of FTPs (try grouping: "section(topdir),firstchar(source),source"). * [curses] Add 'New Source Package View' to Views menu (thanks Thadeu Lima de Souza Cascardo, see #497206), with some updates and minor modifications, like more complex grouping. The menu entry was removed before applying the patch attached to that bug report, see comment to the commit: 66adb6c872f7b49279da43601e7d3ff199996c1a * [curses] Use Yellow/Brown colour to highlight downgrades (Closes: #584510) Previously downgrades were not highlighted, so it was bit confusing/odd and it looked as if nothing would be done with the package. There are not many colours "left" in the limited choice, only Yellow/Brown (same as downloads) and Blue. Since Blue is used in the menus, Yellow/Brown is probably better; Downgrades and Downloads are never on the same screen. * Log file now contains version information for most actions (Closes: #222585) Previously it only contained version information for upgrades and downgrades, both current (old) and candidate (new) versions. Now most actions show version information, either the candidate version (e.g. install, auto install) or current version (remove, hold, ...). * Save selection state of packages to dpkg database (Closes: #137771) Aptitude maintains a separate database (in "pkgstates" file), with a field Dselect-State to keep track of its selected state in dpkg ("status" file). This is updated upon start of aptitude or some operations, so aptitude is aware of changes of this selected state made by other tools; but it was never updated back to dpkg after the selection state changed in aptitude. As a result, for example packages set "hold" in aptitude were then not respected by other tools which do use dpkg selected states. * Consider packages "unknown" to dpkg as not to install (Closes: #328616) While aptitude tried to take into account the selected state of packages in dpkg, the ones not present (with state "unknown", meaning never installed, or purged) were ignored in aptitude, and the internal selected state was used. This was causing problems like not respecting actions taken with other tools (see #328616 and duplicates) -- installing with aptitude and then purging with either dpkg or apt (or higher level applications, like puppet), meant that in the next run of aptitude it would try to install the package again. We now save the selected states of packages from aptitude to dpkg (another feature added in this release), so the selected states are synchronised. If it's "unknown" to dpkg, we will assume that other tools uninstalled it in between aptitude runs, and we will keep it uninstalled. The behaviour of scheduled aptitude actions will be preserved: if a package is selected to install but is not actually installed in the session, aptitude saves the selection in dpkg as "install", so it will be attemped to install in the next aptitude runs (or other tools honouring the selected state in dpkg). Or if a installed package is marked to purge with "aptitude --schedule-only purge PKG", "dpkg --purge --pending" does honour the request. - Bug fixes: * Disable strings gathered for translation (Closes: #790568) * Change to try to make updated status stats more clear (Closes: #458351) * [cmdline] Correctly print when a package should be purged (Closes: #798286) * [curses] With the option of mini-buffer prompts enabled, cancellation of the "Quit confirmation" with 'Escape' or 'Ctrl-g' does not cause the dialog/prompt to stop working (Closes: #607118) * [cmdline] Warn when package cannot be marked/unmarked auto (Closes: #498240) * Install signal handlers and do clean-up when terminating abruptly (Closes: #558352) * [curses] Set better ordering for groups of Preview Actions (Closes: #434352) - User visible changes: * Change error string of source_policy_parser to be the same as the rest * [curses] Improve titles and descriptions of views - Internal changes: * Improve retrieval of source package information with (upcoming) APT 1.1. Thanks to David Kalnischkies for the suggestion and implementation advice. Information about source package and version will be in the binary cache, and available via the VerIterator (ver.SourcePkgName() and ver.SourceVerStr()); and it is much faster than parsing the package record. Additionally, retrieving the source package name in that way defaults to the binary package name if no Source record is explicit, so there is no need to check if it's empty. It is not used in many places in the aptitude code (about 6 files and 10 different functions/places), but this speeds-up and simplifies the code considerably where it was used. * New helper dpkg_selections.{h,cc}, to help to manage saving selections to the dpkg database (--set-selections) - Documentation: * Guide/User's Manual: * Remove grouping policy "hier" from doc/??/aptitude.xml (the categorial/hierarchical browser features were removed in version 0.7) - Translation updates: * Russian by Lev Lamberov (Closes: #798485, #769428) [2015-09-06] Version 0.7.1 - Bug fixes: * Fix for wrong reported size of deleted files with 'autoclean' on 32-bit arches (Closes: #499872) * Be explicit when packages are already installed, cannot be removed/purged because they are not installed, and similar (Closes: #421469). It was changed to work in this way a few years ago in #258913 due to the installer, and required to pass verbose flag to print these messages, but this was caused at least 3 bug reports (merged) to be explicit, and aptitude prints many other messages even ignoring these lines. It can be made to be quiet with another flag, if needed. * Check string formats for columns in both curses and command line interface, and forbid characters which can lead to cwidget assertions and program crashes. These strings can be specified in the the command line for "search" or "versions" with -F or configuration options (and they can be made permanent by mistake, causing crashes upon start, as the bug reports prove). (Closes: #497129, #532662). - Documentation: * Quick help (help.txt): * Document 'l' (Limit Display) in quick help (Closes: #514541) * Improve description of '_' (Purge) (Closes: #495326) * Improve the description of the search shortcut, and point more clearly to search patterns in the full documentation * man page: * Fix example of '-F' with current and candidate versions (Closes: #752517) * Guide/User's Manual: * Upgrade doc/??/aptitude.xml documents from DocBook XML 4.2 to 4.5 * Use XHTML 1.0 rather than HTML 4.01, at least to have something from this decade/century. The resulting documentation still does not fully validate (see #496724), I think that due to bugs/problems of docbook-xsl conversions using invalid elements in the different standards (tried HTML 4.0, 4.01, XHTML 1.0 and 1.1). * Generate 'alt' tag for images in HTML documentation (see #496724) * In "Customizing the package list", fix order of variables ::UI::Package-Status-Format and ::UI::Package-Header-Format and their explanations - User visible changes: * Slightly change description of shortcut 'g' in title bar (Closes: #247487) - Internal changes: * search_graph.h, fix problem with code which was working for many years (with boost 1.55 and many versions before), but now the straightforward option to convert boost::optional to bool does not work with gcc-5.2 and boost-1.58, so use .is_initialized() instead. * solution_item.cc, copy string of package name coming from libapt because it does not guarantee that the string is valid for all of the time span that we need it, to not show garbage that it happens sometimes (Closes: #795150, #795028) * match.cc, pkg_info_screen.cc, reason_fragment.cc: remove and modify code relying on pkg.Section() (from apt PkgIterator), it is deprecated code and it will get removed soon according to the apt team (Versions have Sections, Package in abstract doesn't have unique Sections) * aptitude_resolver.cc: use std::unique_ptr instead of deprecated SPtrArray * Add compatibility for APT 1.1: * Create DepIterator values from Dependency pointers on-the-fly, where required * Adjust AcqQueuedFile() for changed signatures of overriden methods * Improving C++11 compatibilty mode: * Convert std::auto_ptr (deprecated in C++11) to unique_ptr * Convert boost::scoped_ptr (deprecated in C++11) to std::unique_ptr * Convert instances of boost::shared_ptr, make_shared, and weak_ptr to be explicitly prefixed by boost::, otherwise they clash or have problems compiling with the new C++11 std::shared_ptr and analogues (many were not converted in the previous version). * Remove unused function serialize_pattern_list() * Remove unused function wcsempty() * Improvements of the autoconf/automake infrastructure. * No more generate a dummy ChangeLog file just to please automake. * Clean up some more leftover files in "make maintainer-clean" * Use 'cp -u' when copying gtest and gmock, so they don't need to be recompiled every time - Translation updates: * Swedish by Josef Andersson (Closes: #792911) * Fix typos/syntax oddities and translation issues, thanks Jens Seidel (Closes: #381702) [2015-07-25] Version 0.7 - Changes in the versioning scheme and changelog: * From now on Aptitude's version numbers follow loosely the Semantic Version specification (http://semver.org/), with a few exceptions: * The leading zero is kept for now. The first digit after "0." is currently considered being the major version. * Trailing zeros may be omitted, i.e. 0.7 is the same as 0.7.0.0 with 7 being the major version. * Dates in the NEWS file are now consistently written in an unambiguous and more international way. - Feature removals: * The content-wise outdated and not-multiarch-compatible categorical browser has been removed. The Debtags browser has replaced most of its functionality already. Local package groups are no more available, though. (Closes: #686124 and its duplicates as well as #686142 and #686646 as well as LP: #1082198) The removal discussion has taken place in #686646. - Bug fixes: * Fix reported installed size change when installing/deinstalling packages which are bigger than 2 GiB (int type size 32 bits). Thanks Benny Baumann and Jason Rhinelander for the report and info, and Sven Joachim for the patch. (Closes: #759769) - Internal changes: * Use automake option subdir-objects to be able to use current automake versions. * Copy in configure.ac from /usr/src/gtest/src/gtest-all.cc and /usr/src/gmock/src/gmock-all.cc to subdir tests/ -- otherwise, automake (with subdir-objects option recently added) tried to build in /usr/src/gtest/ and ../gmock/, and then failed to link * Remove forward declaration of STL templated classes and containers (basic_string, pair, vector...), use regular includes instead. Due to internal changes of GCC-5 or libsdtc++ related with C++11 compatibility (std::string and std::list had to change), the code failed to compile, and there is not enough justification to "forward-include" in this case (it has to be done copying declarations by hand, it's prone to errors when things are updated in the backstage and not very elegant). * Changes to allow to compile with C++11 compatibilty mode (prompted by changes in the toolchain in Debian): * Convert all instances of boost::make_shared and boost::shared_ptr to be explicitly prefixed by boost::, otherwise they clash with the new C++11 std::shared_ptr. * Convert some instances of boost::shared_ptr and boost::make_shared to the corresponding versions now in the standard library. The reason for this is that it failed to compile straight away with the version in unstable (boost-1.55) in some sections of the code (e.g. problemresolver, sqlite, file_cache), and because since it is now part of the standard, there is little reason to use Boost for that. The rest of cases not covered in this version shall be converted later on. * Several other minor changes to allow to compile with C++11. * Remove duplicated method in sqlite.* files (below). There is another method with the same signature but with "const std::string&", and "const char*" implicitly converts to use that one if there is not one available of the same type. std::shared_ptr statement::prepare(db &parent, const char *sql) * Add top-level update-po target for not needing to remember in which directories the update-po target needs to be called. - Documentation: * Previously missing trigger-related package states are now documented. (Closes: #770073, #771305, #771704) * Minesweeper: Fixed save/load keybindings in help. (Closes: #736934) - Translation updates: * Czech by Miroslav Kure (Closes: #758226) * Russian by Yuri Kozlov (Closes: #760812) * German man page translation by Mario Blättermann and Jens Seidel (Closes: #762340) * Catalan by Orestes Mas (Closes: #771618) * French documentation translation by Cédric Boutillier (Closes: #773807) * Italian documentation translation by Beatrice Torracca (Closes: #776706) * Dutch by Frans Spiesschaert (Closes: #778425) [2014-06-02] Version 0.6.11 - Bug fixes: * Load package tags from APT if debtags database is not available. (Closes: #501732) * [cmdline]: Use arch-qualified names for virtual packages. * Empty environment variables are treated as though they are not set. * Remove -Werror from default compiler flags. (Closes: #746824) - Internal changes: * Access apt-xapian-index directly. Index path can be configured by setting Apt-Xapian-Index::Index in apt.conf(5). * Standardize on a single package tags interface. * No longer use libept. (Closes: #504153, #677551) - Translation updates: * Italian docs. Thanks to Beatrice Torracca. (Closes: #741875) * Portuguese (Closes: #748141) [2014-02-10] Version 0.6.10 Note that 0.6.9 was an experimental release. Most of the changes it contained are temporarily reverted, but they will return. - New features: * New sort policy 'installsizechange', for sorting by change in installed size of each package when the desired actions are applied (Closes: #676075) - Bug fixes: * Fix segfault and show the correct changelog when executing "aptitude changelog ", and no binary packages in that source package match the source package name (Closes: #723821) * Use new URL to grab changelogs, redirections were set up so it is not currently a problem, but the new location seems to be the canonical place for metadata (Closes: #708345) * Fix incorrect results with some sorting policies (Closes: #720750) - Internal changes: * Change INCLUDES to AM_CPPFLAGS in automake files to avoid warning: "'INCLUDES' is the old name for 'AM_CPPFLAGS' (or '*_CPPFLAGS')" - Documentation: * Update manual to refer to canonical git repository. - Translation updates: * Improve and explain some translation messages, thanks helix84@centrum.sk (Closes: #537858). [2014-01-26] Version 0.6.8.4 - New features: * Get changelogs also when package origin is from "Debian Backports" and not only "Debian", thanks Gerfried Fuchs (Closes: #714619) * New sort policy 'debsize' for sorting by package size, thanks Marcel Partap * Add "Reinstall (shortcut L)" to the package menu (Closes: #633788) * Add descriptions for sections introduced in 2011 (education, introspection, metapackages) (Closes: #674681) - Bug fixes: * Do not group packages as tasks based on their name, rather, use their section (Closes: #679602) * Change AC_ARG_ENABLE(package-state-loc, ...) for AC_ARG_WITH(package-state-loc, ...), and the same for "lock-loc", to fix inconsistency/bug in the configure file. Thanks Jeremy Laine (Closes: #399757) * Correct slightly the error message when the invocation of dpkg fails to perform an operation (Closes: #348758) - Documentation: * Improve description and fix typo for the "show" command in man page, thanks Regid Ichira (Closes: #692173) - Translation updates: * Fix inconsistent German translations (Closes: #720186) * Fix important error in translation in Spanish manpage, thanks Martintxo (Closes: #727815) * Fix error in Spanish translation, thanks Manolo Díaz (Closes: #716669) [2014-01-19] Version 0.6.8.3 - Bug fixes: * Avoid segfault with some status line formats (Closes: #714186) * src/pkg_columnizer.cc (setup_column): Avoid a crash by checking that the current item is really a package before accessing package properties. - Build system changes: * Fix FTBFS with g++-4.8 by removing unused typedefs (Closes: #701243) * src/generic/apt/aptitude_resolver.universe.cc * src/generic/apt/resolver_manager.cc * src/generic/problemresolver/choice.h * Support Boost 1.53 (Closes: #710208) * configure.ac: Determine the return type of boost::make_shared. * src/generic/util/sqlite.h: Use that information. * src/generic/util/sqlite.h (statement, blob): Also befriend the new boost::make_shared prototype. Some better solution may be possible at a later point, this is enough for now. * src/cmdline/mocks/terminal.h: Remove predefinition of boost::make_shared that conflicts with Boost 1.53. * Update support for automake and similar tools (autogen.sh) * Do not restrict them to -1.11, use default instead * Add --copy to automake, otherwise creates symbolic links - Internal changes: * task_list is not extern * src/generic/apt/tasks.h: * src/pkg_grouppolicy.cc: Variable task_list is only used within tasks.cc, and need not be extern. - Documentation: * Replace solitary use of "visual mode" with "visual interface" in doc/en/aptitude.xml * Typo in manual: 'Apt::AutoRemove::InstallRecommends' should be 'RecommendsImportant' in doc/en/manpage.xml - Translation updates: * Russian translation of documentation. Thanks to Lev Lamberov and the Russian localization team. * Refresh of .pot and .po files with respect to current source * Some config item strings should be the same width * po/*.po: Use whitespace to adjust the widths where they differ. * src/apt_config_treeitems.cc: Add comment for translators. * German, by Benjamin Weis and Holger Wansing (Closes: #729329) * Japanese (Closes: #728115) * Russian by Yuri Kozlov (Closes: #725280) * Asturian by Inigo Varela (Closes: #698512) * Danish by Morten Bo Johansen (Closes: #700885) * French by David Prévot (Closes: #693323) * Fixed error close to line 3048 due to mismatch in parameters [2012-09-07] Version 0.6.8.2 "Say 'what' again" - New features: * [all]: Support for matching architectures using specification strings and wildcards (for example, "linux-any"). See Debian Policy section 11.1 "Architecture specification strings" for more details. - Minor bugs: * [cmdline]: Fix typo which prevented installing tasks using the apt-get compatible syntax ("aptitude install gnome-desktop^"). * [cmdline]: Do not assume that APT::Default-Release always provides the candidate version of a package, instead use the policy defined by apt_preferences(5). (Closes: #587775, #631464, #686316) * [all]: Download changelogs from APT::Changelogs::Server instead of a fixed location, and replace very ancient and disused option for Aptitude::Changelog-URL-Template. (Closes: #687239) - Documentation: * Update for default value of APT::AutoRemove::SuggestsImportant which is "true" since apt 0.8.15.3. (Closes: #685310) * Correctly reference APT::AutoRemove::RecommendsImportant and SuggestsImportant which do not contain any hyphens. (Closes: #579071) - Translation updates: * Czech (Closes: #681693) * French (Closes: #680040, #685084) * Polish (Closes: #678153) * Russian (Closes: #686016) * Spanish (Closes: #686752) [2012-08-25] Version 0.6.8.1 "Don hat" - Crashes and serious errors: * Multi-arch update for the problem resolver: - handle conflicts without removing all foreign-arch packages (Closes: #672340) (LP: #831768) - correctly recognise dependencies solved by Multi-Arch: foreign packages (LP: #968412) - Minor bugs: * [curses]: Use dpkg package names when calling dpkg (Closes: #680333) - Internal changes: * Add (temporary) helpers for interacting with dpkg. [2012-05-11] Version 0.6.8 "Never need a reason, never need a rhyme" This version of aptitude requires apt 0.9 and will not build against earlier releases. - Crashes and serious errors: * [all]: Fix build with g++ 4.7 Thanks Adrian Lang for the patch (Closes: #672335) - New features: * [all]: Support really big files in the download system This follows recent changes in apt which added support for files greater than several gigabytes in size -- using the 'unsigned long long' type to store the file size. * [all]: Update to tasks support: task packages, multi-arch, syntax Task packages (introduced with tasksel 3.0) are meta-packages which define the dependencies of tasks. The packages themselves have always worked but the 'tasks' grouping policy and '?task' search term did not support them. This update corrects for this. As a result of this change to tasksel all Debian tasks now function exactly like meta-packages. (Closes: #382631) The syntax for installing tasks from the command line has been updated. It now supports specifying an arch and requires the same syntax as apt-utils ('^' must be the last part of the name). Examples: # aptitude install gnome-desktop^ # aptitude install ssh-server^:armel This avoids ambiguity that may arise when a task and package have the same name. - Minor bugs: * [all]: Changelog downloading is restored (Closes: #669569, LP: #824708) * [cmdline]: Exit with non-zero status when a package run is aborted. (Closes: #293008) * Restore candidate version on undo, keep Previously, undo incorrectly restored the candidate to be currently installed version. Keep did not restore it at all. (Closes: #529403) - Cosmetic and UI bugs: * [curses]: Adjust default widths of localized columns: broken_count, downloadsize (Closes: #674045) * [all]: Fold some near-duplicate strings * [curses]: Update package views after 'Cancel pending actions' (Closes: #595753) - Build system changes: * Remove redundancy in doc build system Lots of code was duplicated between makefiles for the various translations. Much of this is now in common rules files making it easier to add new translations. - Documentation: * [doc]: Add doctype headers to html docs. * Italian translation (Closes: #674664) * Fix minor errors in documentation. Thanks to Beatrice Torracca (Closes: #674675) * Fix groff warning in manpage (Closes: #675085) * Fix encoding problems with README files * Install some localized README files which were previously left out. - Translation updates: * French (Closes: #673827, LP: #642840) * German (Closes: #672467, #668875) * Simplified Chinese [2012-05-02] Version 0.6.7 "I said 'Step pause turn pause pivot step step'" - Crashes and serious errors: * [cmdline]: Avoid dpkg and infinite loop in download-only mode (Closes: #629266) (LP: #975793) * [cmdline]: Check for apt_init errors in cmdline_download.cc (Closes: #670379) - Minor bugs: * [all]: Fix various locking issues: - 'install' actions lock /var/cache/apt/archives Other tasks which access that directory (such as 'apt-get clean') will now fail if aptitude is downloading or installing packages. (Closes: #370381) * [all]: Actually use aptitude::Logging::Levels (Closes: #654344) - Internal changes: * [all]: No longer use ListUpdate (from libapt-pkg) to update package lists. Instead, restore the previous code and update it to include the semantics from ListUpdate which were missing. ListUpdate was too coarse to suit download_update_manager properly even though it more-or-less worked. * [all]: Do not call debtags on list update. This was a rather unsightly kludge is better suited as a script for APT::Update::Post-Invoke. * [all]: Destroy download_threads when done with them. The objects were previously left undestroyed, consuming resources and keeping, e.g., locks open longer than needed. - Cosmetic and UI bugs: * [curses]: Apply patch from Ubuntu: don't mention 'su' in 'Become root'. * [cmdline]: Use arch-qualified names in 'Provided by' for virtual packages (LP: #972847) * [curses]: Preview screen ignores Pkg-Display-Limit Previously it would apply that limit if UI::Preview-Limit was not set. This was causing confusion and it is more useful to have these as two clearly distinct settings. (LP: #381507) - Build system changes: * [all]: Include build of gtest when building gmock. (Closes: #670403) * [all]: Fix check_boost.sh for out-of-tree builds - Documentation: * [doc]: Apply patch from Ubuntu: update Finnish manpage to point to doc package. - New translations: * Croatian (Closes: #667934) [2012-03-17] Version 0.6.6 "There is not one you, there are many." This version of aptitude brings in basic multi-arch support. Similar to APT, foreign-arch packages are displayed using their full name, including architecture, for example: 'libc6:armel'. This is intended to provide some support for users of multi-arch systems without causing huge breakage. For users of single-arch systems the changes are limited to displaying the full names of all packages in log files and internal messages. CAUTION: The problem resolver is not yet updated to handle multi-arch properly. Do not rely on it to make intelligent suggestions at this stage, instead, use the architecture information to manually check and resolve any problems. - Removed features: * [gtk]: Remove menu entry for minesweeper. This was causing security concerns by potentially invoking an external program with root privledges. (Closes: #552522) - Security fixes: - Crashes and serious errors: * [cmdline]: List update errors are reported. (Closes: #451137) * [all]: Fix '?action(upgrade)' and others (were finding kept packages instead). Thanks to Rogier for the patch (Closes: #603862) * [curses]: Hack to prevent debug messages dumping to the screen (Closes: #651748) - New features: * [all]: Basic multi-arch support as discussed above. - Use arch-qualified names (full names) in most places. - Sorting by name considers full names (i.e. architecture). - Show architecture and multi-arch fields in package info. - Store package states uniquely by name *and* architecture. - Add related search terms: * ?architecture(architecture), ~rarchitecture Select packages for the given architecture (such as "amd64", or "all"). Special values: "native" and "foreign". * ?multiarch(multiarch) Select packages with a multi-arch capability of multiarch (that is, either "foreign", "same", "allowed", or "none"). - Default search (with no explicit term) now accepts "name:arch" patterns which match "?exact-name(name)?architecture(arch)". - Add architecture grouping policy. (Closes: #659079, #661744) (LP: #454941, #845136, #884945, #904486) * [curses]: Use libraries ("liba", "libb", etc.) with 'firstchar' grouping policy. Removes crowding of the "l" group. - Minor bugs: * [all]: List update now runs APT::Update hooks. (Closes: #476399) * [cmdline]: Actually silence progress messages with '-q2'. (Closes: #141719) * [all]: Fix various locking issues: - '[auto]clean' locks /var/cache/apt/archives; - 'update' locks /var/lib/apt/lists; (Closes: #653479) * [all]: Fix 'priority' cost level in resolver. The values should have been negated but were not. Thanks to Modestas Vainius for this one. (Closes: #608786) * [all]: Make ~VCURRENT etc. behave like ?version(CURRENT). A latent bug was preventing the special values from being handled by the short form. - Internal changes: * [all]: Update package lists using ListUpdate (from libapt-pkg). Now the process is identical to apt-get and other tools, reporting errors more correctly and running the APT::Update hooks. * [all]: Refactor various APT-related comparison functors which were spread all over the place. There is now a single implementation of each which resides sanely in apt.h. * [curses]: Refactor duplicated make_{info,dep,ver}_screen + insert code. Three new functions replace all standard uses of the previous code: show_info_screen, etc. - Cosmetic and UI bugs: * [cmdline]: Fix some help and prompt strings which wrongly included localized names for commands such as 'why'. (Closes: #486615) * [curses]: Change ordering of top-level sections to match the order in aptitude::Sections::Top-Sections. The default order is now: main, contrib, non-free. (Closes: #181997) * [curses]: Improve error messages when parsing grouping policies to use the user-visible names, rather than the internal names. (Closes: #231595) * [all]: Change default width of %D, %I, %o, and %Z. Thanks to Michał Kułach for this one. (Closes: #599209) (LP: #741417) * [all]: Change default width of %V, %v to 14. This greatly increases the number of versions which will display completely. (Closes: #624542) * [cmdline]: Use nice formatting for both parts of the download progress display. Previously the 'downloaded' size was being reported in bytes; now it uses larger units as appropriate to be more readable. * [cmdline]: 'versions' now groups it's results by 'package:arch' instead of 'package'. * [all]: Cleared up grouping, sorting policy error messages. (Closes: #231595) - Build system changes: * [all]: Add configure option '--disable-tests'. * [all]: Don't check for libvte if GTK+ is disabled. * [all]: Apply patch from Ubuntu to fix build -D_FORTIFY_SOURCE=2 errors. Thanks to Colin Watson and Michael Vogt (Closes: #497539) * [all]: Avoid most library and header checks if --disable-aptitude. * [all]: GTK+ frontend is disabled by default. * [all]: Enabling GTK+ or Qt is an error if libraries are missing. - Documentation: * [doc]: Minor correction to the user's manual. (Closes: #365485) * [doc]: Clean recent changes to man page description of 'forbid-version'. - Translation updates: * Brazilian Portuguese (Closes: #483964) * Hungarian (Closes: #592468) * Polish (Closes: #659803) * Russian (Closes: #662624) [2012-02-06] Version 0.6.5 "Splat!" - Removed features: * [cmdline]: Remove broken safe-upgrade-only options. Specifically, Safe-Upgrade::No-New-Installs and Safe-Upgrade::Show-Resolver-Actions. These options were documented but have never worked. A quick skim didn't turn up any bugs filed about them, so I guess no-one ever tried them. Removed them. - Minor bugs: * [all]: Do not translate program options in Catalan translation. (Closes: #462558) * [all]: Do not translate program options in Simplified Chinese translation. (Closes: #563132) - Internal changes: * [gtk]: Hopefully fix the use of VTE in the GTK+ frontend. * [all]: Don't run file in aptitude-run-state-bundle. Thanks to Sven Joachim for the patch. (Closes: #635572) - Cosmetic and UI bugs: * [curses]: Handles 'Enhances' relations in preview. (Closes: #583201) * [cmdline]: Fix resolver help for next/previous. (Closes: #495046) * [curses]: Boldness of Tags and User-Tags fields. (Closes: #652360) * [curses]: Wrap long description of boolean_config_treeitems. (Closes: #653120) * [cmdline]: Add versions, newlines to --help. (Closes: #581597, #604392) * [all]: Fix prompt strings. (Closes: #652419) * [cmdline]: Typos in output of version, usage. - Build system changes: * [all]: Clean up the use of AC_COMPILE_IFELSE and AC_LINK_IFELSE to fix a bunch of autoconf warnings. * [all]: Modify autoconf/automake scripts to handle gmock changes. * [all]: Silence automake warnings about GNU extensions. - Documentation: * [doc]: Cleaned up the documentation of --new-new-upgrades and --no-new-installs. (Closes: #568297) * [doc]: Minor fixes in documentation and changelog. (Closes: #403220) * [doc]: More precise wording suggested by jidanni in the manpage. (Closes: #542293) * [doc]: Switch Polish manpages to po4a. Thanks to Michał Kułach. (Closes: #656820) * [doc]: Fix link in users manual. (Closes: #397038) * [doc]: Fix missing entity refs and filename elements. (Closes: #651289) * [doc]: Update development details in the user's reference. (Closes: #651288) * [doc]: Fix description of Aptitude::Sections::Top-Sections. (Closes: #651661) * [doc]: Mention /etc/apt/apt.conf.d/* in Configuration file. (Closes: #497138) * [doc]: Fix manpage text for safe-resolver. Thanks to Dan Jacobson (Closes: #511731, #568297) * [doc]: Update description of forbid-version. (Closes: #511365) * [doc]: Fix typos in manpage. Thanks to A. Costa for reporting them (Closes: #644651) * [doc]: List items wrongly nested in the document. Thanks Kobayashi Noritada (Closes: #365615) * [doc]: Update dead URL for regular expressions in the manual, thanks Diederik de Haas (Closes: #581508) * [doc]: Properly document 'Q' and 'q' keys for user interface in the Help file. Thanks Sebastian Kapfer and Stéphane Aulery. (Closes: #242269) - Translation updates: * Calatan (Closes: #332592) * Danish (Closes: #608593) * Dutch (Closes: #601672) * French (Closes: #631759) * German (Closes: #603589) * Greek (Closes: #606871) * Hungarian (Closes: #548609) * Japanese (Closes: #550451) * Polish (Closes: #657174) * Portuguese (Closes: #577995) * Russian (Closes: #559664) * Simplified Chinese (Closes: #594526, #566367) * Slovak (Closes: #559288, #653056) * Spanish (Closes: #564445) [2011-04-02] Version 0.6.4 "Spring is sprung" This version of aptitude requires apt 0.8 and will not build against earlier releases. - Security fixes: * [curses]: Fixed an arbitrary overwrite triggered by saving changes in the hierarchy editor, if the current user's home directory couldn't be determined. (Closes: #612034) - Crashes and serious errors: * [all]: Don't crash when apt-xapian-index isn't installed. (Closes: #588089) - Interprets "~" as "?name(~)" instead of "?term(~)" - If the Xapian index isn't available, attempts to mimic its behavior to a first approximation: not perfectly, but enough to make ?term and ?term-prefix do something sensible. * [all]: Restore aptitude's old treatment of the auto flag, meaning that instead of clearing the flag when the user asks to upgrade a package, we leave it set. Worked around a change in apt's behavior by forcibly preserving the auto flag's setting regardless of what pkgDepCache::MarkInstall thinks is right this week. (Closes: #622719) - New features: * [curses]: Add a "source" grouping policy that groups packages by their source package name. Thanks to Thadeu Lima de Sourza Cascardo for the patch, sorry it took so long to apply! (Closes: #497206) - Minor bugs: * [all]: Fixed the debug logging code to output TRACE-level messages. * [all]: Fixed the debug logging code to properly allow log levels to be disabled. * [all]: Fixed the resolver cost levels for installs and upgrades. - Internal changes: * [all]: Include the current time and thread ID in all debug log messages. * [cmdline]: Refactor terminal handling to simplify some of the tests. * [qt]: Added the beginning of a Qt frontend, thanks to SoC 2010 student Piotr Galiszewski. Not really in a usable state yet. * [all]: Fixed build errors with gcc 4.6 (Closes: #624948) - Cosmetic and UI bugs: * [cmdline]: Overhaul how download progress is displayed at the command-line. Fixes problems on large terminals and should provide less glitchy behavior in general. - Build system changes: * [all] Use $(srcdir) whenever $(wildcard) is used, so wildcarded files get picked up even when the build directory is not the source directory. * [all] Build src/generic/util before src/generic/apt (which depends on it). * [all] Multiple fixes to get VPATH builds working properly. * [doc]: Added a doc-fo target that builds (very rough) PDF documentation if you have fop installed. * [gtk] If the GTK+ frontend isn't built, don't try to build tests that require it. [2010-06-18] Version 0.6.3 "Sunlight, moonshadow" This version of aptitude fixes some crashes that occurred when exiting the program and overhauls how progress is displayed from the command-line. aptitude now requires google-test and google-mock; it will work with newer versions of libept and older versions of boost; it no longer requires log4cxx. - New features: * [cmdline] Modified the command-line progress counters so that they're erased when they complete instead of leaving cruft behind. The old behavior and appearance can be recovered by Aptitude::CmdLine::Progress::Retain-Completed and Aptitude::CmdLine::Progress::Percent-On-Right to "true". - Crashes and serious errors: * [all] Ditching log4cxx will hopefully fix some of the crashes on exit that people were seeing. - Cosmetic and UI bugs: * [cmdline, curses] Fixed the display of the archives associated with a version so that duplicated archive names are never displayed. - Internal changes: * [gtk] Started working on a redesign of the GTK+ interface, both internal and external. The internal half involves refactoring the code to better separate logic and presentation and to improve testability and code reuse. The external half adds the groundwork for a redesign of how the UI is arranged at the top level. This code is very incomplete and currently disabled by default. * [cmdline] The progress changes noted above included some refactoring to make the command-line interface easier to unit test. In particular, I created a terminal abstraction that could be stubbed out for the purposes of testing. Currently this is only used for the progress code, but it's likely to become more widely used as time goes on. - Changes for build dependencies: * [all] Stopped using log4cxx in favor of a homegrown module. We're really only using it for debug output, so its most powerful features aren't useful for aptitude. At the same time, it appears to be behind some of the crashes on exit that users have reported, and it's never really integrated well into the program (the initialization process has always been a hack). The new code is much better suited to aptitude's requirements. * [all] Added appropriate configure checks and conditionalization to support building against either the current ept release or the upcoming release. (Closes: #581009) * [all] Write a proper configure check to detect the calling convention used by boost::fusion::fold(); don't require Boost 1.42 any more. - Build system changes: * [all] Added a build system based on scons. scons supports fully parallel builds and variant builds without reconfiguration, which should speed up the build process significantly on multi-core machines. * [all] Switch from automake 1.9 to automake 1.11. [2010-04-21] Version 0.6.2.1 "I didn't know it was a real zebra!" - Crashes and serious errors: * Fixed a crash caused by being insufficiently careful about NULL pointers in the resolver. (Closes: #578344) * Boost 1.42 changed the order of arguments to fusion::fold(), breaking aptitude. As an interim measure, I have modified configure to require 1.42 or higher, and fixed the argument order. The correct long-term fix here is to do a feature test to see which order fold expects its arguments in. - Backwards compatibility fixes: * Restored the "conflict" safety level/tier. * Fixed a bug that meant that the old configuration option names (*-Tier instead of *-Level) could only be used to increase safety levels, not to decrease them. - Internal changes: * Change the incremental_expression subsystem of the resolver to be fully robust against NULL pointers and write unit tests for it. * Reduced the number of explicit dependencies on log4cxx (using indirection), in case I decide to resolve the random crashes it causes by switching to some other framework or rolling my own. [2010-04-18] Version 0.6.2 "A costly proposition" - New features: + [all] Resolver tiers are dead, long live resolver costs. The resolver now supports a flexible system of multi-leveled costs, in place of the fairly rigid tiers of previous releases. The default behavior emulates how tiers worked, but the resolver can also be configured with custom cost vectors by setting Aptitude::ProblemResolver::SolutionCost to something like this: canceled-actions + 2*removals, ignored-recommends, installs That says to minimize the number of packages kept at their current version plus twice the number of removals (i.e., a removal counts for two keeps). Within that group, ties are broken by looking at the number of Recommends that were ignored, and within that group, ties are broken by looking at the number of new packages the resolver wants to install. In addition to custom cost vectors, resolver hints can be used to create completely custom costs. See the reference manual for full details. This is not optimized and I expect that pushing it to the limits will show off all sorts of exponential explosions. Have fun! + [cmdline] Implemented an "aptitude versions" command that I've been kicking around for a while, to display and search for individual package versions. Documented in the manpage, but here are some highlights: $ aptitude versions apt i 0.7.25.3 unstable 500 p 0.7.26~exp3 experimental 1 $ aptitude versions '^apt$ ?installed' Package apt i 0.7.25.3 unstable 500 $ aptitude versions xserver-xorg udev Package udev: i 151-2 100 p 151-3 unstable 500 Package xserver-xorg: i A 1:7.5+3 100 p A 1:7.5+5 unstable 500 $ aptitude versions --group-by=source-package '?name(aptitude)' Source package aptitude: i aptitude 0.6.1.5-2 100 p aptitude 0.6.1.5-3 unstable 500 i aptitude-dbg 0.6.1.5-2 100 p aptitude-dbg 0.6.1.5-3 unstable 500 i aptitude-doc-cs 0.6.1.5-2 100 p aptitude-doc-cs 0.6.1.5-3 unstable 500 (... many more lines of output snipped ...) - Crashes and serious errors: + [cmdline] Don't crash if the user sets a rejection at the internal command-line via package name and version rather than using the new pick-by-numbers UI. (Closes: #567242) + [cmdline] Don't crash if the user asks to see information about a removal. + [cmdline] Don't crash in "aptitude changelog" when the package given on the command-line doesn't have a valid version (Closes: #576318). Thinks to Dmitry Semyonov for the patch. - Cosmetic and UI bugs: + [curses] Don't leave a download progress bar hanging around after downloading a changelog. (Closes: #566205) - Internal changes: + [all] Replaced mktemp() with a custom function doing a similar thing (albeit with slightly better randomness). This eliminates the spurious linker warning I've been ignoring for five years: the linker has no way to know that I only use mktemp() to create names in a mode 0700 directory, which is about the only way to use it safely. [2010-01-20] Version 0.6.1.5 "All around the dependency bush, the monkey chased the weasel..." - Crashes and serious errors: + [all] Changed the default resolver configuration so that removals and "safe" actions are grouped together. Without this change, aptitude was being too conservative and failing to solve simple upgrades. (Closes: #565867) The safe-upgrade resolver is unaffected by this change, since it uses explicit constraints to prevent removals from being installed. - Cosmetic and UI bugs: + [test] Eliminated a spurious error message that was being printed by the cppunit unit tests. + [curses] Fixed the progress display and error reporting for changelog downloads. - Documentation: + [doc] Added a note in the manpage explaining the principles of the Linux command-line to hopefully forestall future bug reports telling me that the manpage for "aptitude search" contradicts the user manual. (Closes: #566003) [2010-01-17] Version 0.6.1.4 "You can't take the sky from me." - New features: + [gtk] The status of each changelog download is now displayed in the space where that changelog will appear. + [cmdline] When aptitude displays a resolver solution, it now displays a number next to each entry in the solution. This number can be used to manipulate the corresponding entry (for instance, to reject it or to view information about it). This makes it much more convenient to use the command-line resolver interactively. - Crashes and serious errors: + [all] Be paranoid about the sign of file offsets (on some architectures they're apparently signed). (Closes: #557212) + [all] Fix various include and compilation errors revealed by the new version of g++ being stricter. (Closes: #531687, #560517) + [all] Use pthread_sigmask() instead of sigprocmask() on startup to avoid problems on architectures where sigprocmask() behaves poorly in multi-threaded processes. (Closes: #559980) + [gtk] Fix a startup crash caused by not properly protecting the call to Glib::thread_init(). (Closes: #555120) - Cosmetic and UI bugs: + [cmdline] Document the flags that are shown after packages in the preview. + [cmdline] Instead of showing broken packages in their own group, show them in the group they would normally appear in but with a "b" flag. It was somewhat confusing that packages which were being installed but had broken dependencies showed up in the "broken" section instead of the "install" section. Packages that are unchanged and have broken dependencies won't show up in the preview any more, but they will show up just below it, in the list of packages with broken dependencies. + [all] Eliminate spurious error messages from log4cxx. (Closes: #557150) - Internal changes: + [all] Download changelogs using the new internal download queue, not the download_manager system. Fixes random crashes at shutdown. + [gtk] Wrote the first draft of some harness code that will underpin a redesign of the main window. Currently this is unused. [2009-11-19] Version 0.6.1.3 "Too many cooks put lots of beans in the soup." - Build system fixes: + [doc] Fixed the Makefile for the Spanish translation of the documentation so that it honors DESTDIR. [2009-11-19] Version 0.6.1.2 "I think I can, I think I can..." - Crashes and serious erroors: + [all] Modified the new temporary file module so that it doesn't rely on the order in which global destructors run. This was causing an abort when the program exited if it was built with optimizations. - Internal changes: + [test] Wrote an additional unit test verifying that the program's temporary directory is really deleted when it exits. [2009-11-19] Version 0.6.1.1 "Again with the bug-writing." - Crashes and serious errors: + [test] Fixed the test code to actually compile against the new temp API (major oops there). + [all] Fixed the temporary directory code to create directories in $TMPDIR rather than ".". [2009-11-18] Version 0.6.1 "Instant toll logging." - New features: + [gtk] Support for automatically downloading screenshots from screenshot.debian.net. Eventually the screenshots will probably show up in a few more places, such as the update preview and package lists (but this will require careful coding to only download screenshots that are visible). - Internal changes: + [all] aptitude now has a generic global download queue. This is actually implemented with the apt download queue object, but it runs all downloads in the background, integrates with aptitude's download cache, and provides support for getting status information on individual downloads and canceling them. This new queue is used to implement screenshot support. + [all] Overhauled how temporary files are created. aptitude will now create only one entry in $TMPDIR, and it will make a stronger effort to delete all its temporary directories when it exits normally. Abnormal exits (i.e., signals) will still leave cruft behind, but at least it's only one file and not dozens per run. - Crashes and serious errors: + [all] Don't support trying to block the foreground thread while dependencies are solved; doing so is very deadlock-prone and was causing deadlocks in practice. + [all] Fix several places where the "why" code dereferenced version iterators without checking whether they were valid, causing a crash. (Closes: #546446) + [gtk] Fix several crashes in the resolver tab caused by trying to generate tooltips using invalid version objects. + [gtk] Properly disconnect signals from the background thread in the dashboard tab, instead of occasionally crashing when one is delivered after they should have been disconnected. + [gtk] Fix a potential deadlock that could have occurred if the main thread posted lots of thunks to itself. - Minor bugs: + [all] The resolver will now ignore recommendations that are already unresolved on the system, rather than trying to fix them every time any dependency needs to be solved. (Closes: #556042) + [curses] Fix becoming root with the "su" protocol. (Closes: #552471) - Cosmetic and UI bugs: + [all] Correct how we decide when the resolver is out of solutions. + [all] Work around a spurious error message generated by apt. (Closes: #526661) + [cmdline] The message that no packages are being installed, removed or upgraded is more accurate now. + [cmdline] Another preview is shown after the resolver runs even if -y was passed on the command-line. + [cmdline] Exit without prompting if the user tells the dependency solver to cancel all of their changes. (Closes: #556866) + [gtk] Fix a typo in one of the resolver tab's tooltips. + [gtk] Fix the resolver tab to only show a new solution when the user explicitly asked for one. [2009-10-25] Version 0.6.0.1 "Who's laughing now?" - Build system fixes: + [all] Actually distribute the auxiliary data for the unit tests. Hopefully this means that the unit tests can actually pass in the distributed tarfile. [2009-10-25] Version 0.6.0 "Stick a fork in it, it's done." - Internal changes: + [all] The file cache now supports storing the last modification time of cached files. Currently this isn't used (changelogs are assumed to be up-to-date anyway), but future code will need this. + [gtk] Factored out the common parts of the background threads that handle changelog downloading into a generic utility class. - Minor bugs: + [all] Load the download cache in apt_init() instead of apt_preinit(). Avoids complaints from log4cxx if there are errors, and also ensures that we have a download cache even if the apt cache was closed and reopened. + [all] Add some missing #includes that showed up when NLS wasn't enabled. - Build system fixes: + [all] Don't include files in POTFILES.in that aren't distributed. Hopefully this will prevent the unnecessary junk that was getting into the Debian diff. [2009-10-12] Version 0.5.9rc4 "Trust me, you don't want to know about the spork." - New features: + [all] The test to see whether a changelog is in aptitude's download cache now takes place in a background thread. + [gtk] When generating the upgrade summary, aptitude caches a parsed version of the portion of each changelog that's displayed. This makes the changelogs show up much faster and with a lot less system load. + [gtk] The second tab in the upgrade view now displays information about the currently selected package, rather than its full changelog. + [gtk] "aptitude --version" prints the versions of GTK+ and GTK-- that it was compiled against. - Crashes and serious errors: + [all] Fix how the background dependency solver thread communicates with the foreground thread to eliminate at least one deadlock. - Cosmetic and UI bugs: + [curses] Don't print a spurious warning if we get --no-gui when the GTK+ frontend was compiled out. (Closes: #550727) - Build system fixes: + [gtk] Make "./configure --enable-gtk" enable GTK+ instead of disabling it. [2009-10-08] Version 0.5.9rc3 "The stars, like little starry things..." - New features: + [cmdline] Package versions can be chosen by codename as well as by archive. (Closes: #547707) + [all] Not exactly a new feature, but reverted from dpkg-parsechangelog to parsechangelog: both of them require extra dependencies, and parsechangelog is faster. (Closes: #546280) - Crashes and serious errors: + [all] Don't crash if the Xapian database fails to load. (Closes: #512998) + [all] Recognize --no-gui even in non-GUI builds, so switching to root can just always pass it. (Closes: #546406) + [all] Don't discard errors after dpkg fails. (Closes: #548879, #548889) + [gtk] Don't crash when displaying the file list of a package that's not installed. - Cosmetic and UI bugs: + [curses] Block SIGWINCH by default so that cwidget can sigwait() on it. (Closes: #547212) + [cmdline] Mention build-dep in "aptitude help". (Closes: #547151) + [cmdline] Consistently use a full-stop at the end of "--help" lines. + [cmdline] Fix an incorrect message that stated that a package was not installable, rather than that it was not going to be installed. + [cmdline] Make --arch-only do what it should instead of the opposite. (Closes: #547266) + [gtk] Make the "hold" command hold packages instead of deleting them. (Closes: #549897) + [gtk] Use a single thread for all changelog downloads, to avoid spawning as many instances of parsechangelog as there are changelogs. I really should move the extraction of changelogs from the cache to a background thread as well; it takes ages to extract every last changelog if there are a lot, and I should be able to start parsing them as soon as the first one is extracted. - Documentation: + [doc] Mention Get-Root-Command in the section on "becoming root". (Closes: #548657) + [doc] Correct the default value of Get-Root-Command to match the code. (Closes: #550057) - Translation updates: + Danish (Closes: #546497) + French + Portuguese (Closes: #522445) + Russian (Closes: #535806) + Spanish [2009-09-09] Version 0.5.9rc2 "Over five served." - Crashes and serious errors: + [all] Fix a bug in the logic to filter package actions so that only actions initiated by apt's greedy resolver are affected, rather than all actions. [2009-09-08] Version 0.5.9rc1 "The only way to win is not to play." First release candidate for 0.6.0. Due to the length of time it's taken me to prepare this and to the new Debian release cycle, I'm releasing 0.6.0 before the GTK+ interface is really complete; it's still in an experimental state. However, several of the other changes in this release should really be rolled into the mainline before the Debian release (particularly the improvements and optimizations to the dependency solver). This version disables the full-text Xapian search by default, although it can still be accessed via ?term(). Version 0.5.9rc1 - New features: + [all] Internal file cache for auxiliary downloads, based on the binary file engine SQLite and the Boost zlib wrapper. This will substantially speed up the GTK+ interface's startup time when it's just downloading the same thirty changelogs it was showing before, and will be even more helpful once aptitude can show screenshots. Hopefully it will also reduce the load on the Debian network infrastructure. + [all] aptitude now uses the new hooks in apt to prevent the greedy resolver from removing packages or breaking holds. (Closes: #177374, #205049, #374353, #376802, #406506, #430816, #434731, #442420, #452589) + [all] aptitude now uses dpkg-parsechangelog to parse changelogs instead of parsechangelog, courtesy of Frank Lichtenheld's patch in 2007. This means that it no longer requires libparse-debianchangelog-perl to show changelogs. + [all] "aptitude update" now honors the configuration options APT::Get::List-Cleanup and APT::List-Cleanup. (Closes: #448958) + [cmdline] "aptitude safe-upgrade" now supports listing extra actions on the command-line, just like "install", "remove", etc. If any actions are listed, then only the given actions are performed (the blanket upgrade is skipped). Packages with no qualifications give a list of the packages that should be upgraded; if they aren't installed or aren't upgradable, they're ignored. + [curses] The commands to clean the package cache and to delete obsolete files now support automatically becoming root. (Closes: #492832) - Crashes and serious errors: + [all] Fix some crashes that occurred when viewing dependency solutions. + [all] Make "aptitude update" correctly clean up the list download directories. (Closes: #507603) + [all] Disable Xapian full-text search by default; it's still available via the ?term pattern. Feedback on this "feature" was uniformly negative; it seems to be flaky and not very useful. Maybe it would be useful with more work, but it won't be ready for Squeeze. (Closes: #527540, #538198, #514625) + [cmdline] Fix a bug that caused lots of extra "unresolved recommendations" to be displayed in dependency solutions. + [curses] Fix a crash on exit that was caused by having dangling boost::flyweight objects when the global destructor was invoked (arguably this is a bug in Boost; we might want to consider moving away from flyweights for this reason). + [curses] The auto-su-to-root code now passes --no-gui, so it should actually work again (rather than popping up a graphical UI). - Cosmetic and UI bugs: + [all] Fix some corner cases in the serialization of search patterns, such as the new "enhances" dependency type. + [curses] Hopefully fix pattern-based grouping in the curses UI. + [curses] Add descriptions of the new ftp sections to the default sections file (thanks to Miroslav Kure). + [gtk] Make more of the UI translatable (thanks to Miroslav Kure). + [gtk] Don't attempt to translate the empty string when displaying the state of a package that's unchanged. - Documentation: + [doc] Document that full-upgrade and safe-upgrade take extra actions on the command-line. (Closes: #268696) + [doc] Document that the "standard" apt options should be used instead of Keep-Suggests and Keep-Recommends. (Closes: #540230) - Translation updates: + Czech (Closes: #545144) + French + Spanish [2009-07-21] Version 0.5.3.1 "Sky without end" - Build system fixes: + Fixed the Spanish documentation Makefile to use DESTDIR. - Translation updates: + Slovak (Closes: #521052) [2009-07-20] Version 0.5.3 "The Aardvark Contingency" This version of aptitude now requires GTK+ 2.16 or greater and GTK-- 2.16 or greater. - Crashes and serious errors: + [gtk] Don't crash when displaying a solution that cancels the installation of a package. (Closes: #525898) - Internal changes: + [all] Rewrote the resolver code to trade memory for time. Large dependency problems should be processed in about 20% of the time that previous versions took. - Cosmetic and UI bugs: + [gtk] Some redesigns to the GTK+ UI (Thanks to Arthur Liu). + [gtk] Display package short descriptions in the resolver tab. + [gtk] Display the search progress bar behind the text of the search term (requires GTK+ 2.16 or greater). - Translation updates: + Asturian (Closes: #521558) + Finnish (Closes: #531582) + Spanish: added translation of the user's manual. [2009-04-26] Version 0.5.2.1 "Albatross!" - Crashes and serious errors: + Don't crash when tabs are closed from the menu or using Ctrl-W. - Internal changes: + [gtk] Always use trampolines (via the main loop's idle handler) to delete tabs, to hopefully avoid bad interactions when a tab is closed as a side effect of some other action. - Cosmetic and UI bugs: + [gtk] Eliminated some warnings that were caused by modifying a tree model while iterating over its selection list. + [curses] Work around some buggy terminals so that the progress bar doesn't get truncated or corrupted. (Closes: #455220) + [cmdline] Change the "last-package" summary mode to "first-package". aptitude prints the package that's selected by this option at the beginning of a chain, so it seems counterintuitive to me to call it the "last" package. + [cmdline] Renamed the configuration option "Aptitude::Why-Display-Mode" to "Aptitude::Show-Summary", to match the command-line flag that it corresponds to (--show-summary). + [cmdline] Removed a stray space in the output of "aptitude why". - Documentation: + [doc] Document --show-summary and Aptitude::CmdLine::Show-Summary. + [doc] Use rsvg instead of Inkscape to convert svg diagrams to png files for the time being. Also cropped rsvg to the drawing boundaries. + [doc] Add search-tier-diagram.svg to the distributed .tar.gz file. - Translation updates: + French [2009-04-22] Version 0.5.2 "Ramparts of the Dawn" This release merges in the work that was done on the lenny branch of the program but not the post-lenny one (mostly translation work). KNOWN BUGS: + Ctrl-W should close a tab, but instead it crashes the program. Do not press Ctrl-W and you will be happier. + When you reject or accept a resolver choice, the corresponding row in the solution might not update to reflect your change. If you switch to another solution and back, it will update correctly. - New features: + [all] aptitude now uses liblog4cxx to provide diagnostic logging of much of the program. + [all] The aptitude dependency solver now supports a "tiered search". Hopefully this will make dependency resolution more predictable and allow more meaningful configuration. In particular, removals are not attempted until all solutions involving only keeps, installs and upgrades are exhausted, and solutions involving versions that aren't the default candidate are deferred until removals are exhausted. See the user's manual for details. Closes: #514820, #514930, #524221, #473296 + [gtk] The resolver tab has been completely overhauled. It now supports all the features of the backend dependency solver, and it has an interface that is much more suitable for a GTK+ program. + [gtk] The dpkg tab now attempts to detect when something is waiting for input, by keeping track of how long it's been since dpkg sent a status message. When input is available, the "view details" button is flashed to hopefully get the user's attention. + [gtk] The procedure for upgrading has been streamlined. aptitude computes an upgrade in the background automatically using the same algorithm that backs "safe-upgrade". The user can either accept the solution or try to fix the remaining dependencies manually. + [cmdline] Added a new command-line option, "--show-summary", to the "why" command-line action. This option causes aptitude to show a brief list of the first package in each dependency chain that would have been displayed. So combined with the "-v" option, you can easily get a list of all the manually installed packages that require a given package. + [gtk] When the user clicks on a package in the dashboard's list of upgrades, the changelog display automatically scrolls to that package. + [gtk] Some common notifications are now marked with the icon of the task they're performing. + [gtk] Initial support for incremental search (currently you have to click the "incremental" toggle button to enable it). + [gtk] Ctrl-PageDown and Ctrl-PageUp, or F7 and F6, now switch to the next or previous tab, respectively. + [gtk] A link now exists below the package description to open all the other packages from the same source package. + [all] The aptitude dependency resolver now continues searching for a few steps after it finds a solution, in the hope that it can find a better one. (Closes: #482825) + [all] The aptitude dependency resolver now adds a bonus to the default apt resolution for a dependency (that is, the first non-virtual package listed). The default bonus is 400 and it may be configured via Aptitude::ProblemResolver::DefaultResolutionScore. + [all] Setting the environment variable APT_ROOTDIR to a directory name will cause aptitude to use the given directory as the root for the purpose of locating apt files. - Crashes and serious errors: + [gtk] Don't crash in the dependency chains tab. (Closes: #514714) + [gtk] Don't crash when the user views the information tab of a virtual package. + [all] Fixed some serious inefficiencies in the algorithm used by safe-upgrade. It's still a bit slow on very large upgrades, but not as ridiculously slow as it was before. + [all] Fixed a bad interaction between the dependency solver and reinstatement of unused packages, which could cause a situation where a "solution" from the solver would actually lead to broken dependencies, due to a conflict declared on the newly reinstated package. (Closes: #522881, #524667) - Cosmetic and UI bugs: + [gtk] Changelogs are now downloaded and parsed in the background. aptitude still sets the download up in the foreground, and this can take a long time if there are lots of available upgrades. + [gtk] A background thread is used to find the packages that match a search pattern and to build the list of matching packages for display. + [gtk] Some work on the download screen, although more is needed. + [curses] Don't warn the user about removing an Essential package if they're just purging the configuration files of one they removed already. (Closes: #513472) + [curses] Don't compute an excessive amount of "why" information every time a package is selected. Should make the curses interface much less sluggish. (Closes: #516296) + [cmdline] Add a missing newline in the error message that "aptitude why" prints if it has no arguments. (Closes: #514676) + [cmdline] Fix some badly formatted messages that are displayed when a build dependency can't be installed: they were missing punctuation and newlines. (Closes: #492615) + [cmdline] Make the warning asking the user to use safe-upgrade instead of upgrade a bit clearer. + [cmdline] Always display [held] next to the state of held packages. + [cmdline] Pre-compute the column sizes needed to display "why", so that columns don't wrap in Stupid Places[tm]. (Closes: #516389) - Minor bugs: + [all] The dependency resolver's scoring of full package replacement was not behaving as intended in the corner case that the package was replaced due to a virtual package it provided, and some other version of the replaced package didn't provide that virtual package. The intent was to give the non-providing package a bonus, but instead the versions that got replaced were getting a bonus (more or less negating the whole point of the full replacement score). + [all] The resolver should now more correctly obey approval constraints set by the user (solutions that were previously rejected and aren't rejected any more can be returned). - Internal changes: + [all] The dependency resolver now supports "hypothetical reasoning": it can calculate a solution to a dependency problem from a starting state that isn't the "live" state of the package cache. This is used, for instance, to compute an upgrade before any packages are marked for upgrade. + [all] Implemented support in the dependency resolver for propagating knowledge about dead-ends up the search tree. Currently disabled as it turned out to greatly increase the memory and CPU overhead of a search without significantly decreasing the number of steps needed. + [all] The dependency resolver now stores solutions as sets of "choices" rather than explicitly maintaining several lists for the different types of choices a solution can contain (for instance, installing a version vs leaving a Recommends unresolved). This makes things a bit cleaner and should make it easier to add new types of choices if that turns out to be useful. + [gtk] The Glade file now stores each tab as a separate top-level widget, which makes it a lot easier to edit and add tabs. - Documentation: + [doc] Improve the documentation of --purge-unused. + [doc] Improved documentation of the resolver (in addition to documenting the new features). - Translation updates: + Asturian (Closes: #518981, #519693) + Danish (Closes: #512384) + German + Italian + Norwegian Bokmål (Closes: #510977) + Spanish (Closes: #517272) + Swedish (Closes: #511238, #514011) [2009-01-18] Version 0.5.1 "One step at a time" - New features: + [all] New match term: ?term-prefix(term). This is like ?term(), but it searches the Xapian database for any term which is an extension of ?term. For instance, ?term-prefix(hour) matches any package that contains the terms "hour", "hourglass", "hourly", etc. + [all] If the user asks for the changelog of a version of a package that's currently installed, aptitude will use the local copy instead of downloading a new copy off the network. If the local copy can't be used for some reason, aptitude will fall back to downloading a copy. + [all] Implemented resolver hints: you can now place options in apt.conf that control how the aptitude resolver treats particular packages. For a full discussion see the user's manual, but here is a flavor: Aptitude::ProblemResolver::Hints { "reject pulseaudio"; // Prevent the "pulseaudio" // package from being installed // by the aptitude resolver. "approve ?name(emacs[0-9]*)"; // Always choose emacsNN over // any alternatives. "300 ?true /unstable"; // Give a 300-point bonus to // any version in the // "unstable" archive. }; Currently these hints do not influence the choices made by the "immediate" dependency resolver (the one that runs as soon as you mark a package for installation). + [gtk] aptitude now parses the dpkg status pipe, so it can show a progress bar while dpkg is running. Closing the terminal tab does not terminate the install; only destroying the progress bar does that. + [gtk] The search entry box's background will turn slightly red if the current text is not a valid search pattern. + [gtk] If the user enters an invalid search pattern into the search entry box, the error will be shown below the box (rather than being shoved onto the apt errors pane). + [gtk] aptitude prompts for confirmation before killing a download or a dpkg process. The download prompt might be superfluous, but considering the possible side-effects of killing off dpkg, that prompt will remain for the foreseeable future. + [gtk] aptitude will print a message to its embedded terminal before and after running dpkg. + [gtk] A drop-down box of package filters is available under the text box where searches are entered. This lets you quickly filter a list of packages by some common criteria (showing only packages that aren't installed). + [gtk] The "automatically installed" flags of packages can be toggled from the Package menu or from buttons in the description pane. + [gtk] Undo -> Undo works. + [gtk] When viewing a dependency solution, you can choose to see the actions in their "logical" order (equivalent to typing "o" in the curses front-end). + [gtk] In the version information tab (the one you get by double-clicking a package), selecting a different version in the list of versions causes information for that version to be shown instead. + [gtk] You can now change which columns are visible in a package list by selecting View -> Edit Columns... or by clicking the "..." column heading. + [gtk] New package list column: "automatically installed". Each value in this column is displayed as a check-box, where the box is checked if the package is automatic. The check-box can't currently be clicked because I'm worried that people would click it by accident while trying to select a package. Most likely the rendering will be changed in the future to look less "clickable". + [gtk] New package list column: "archive". - Crashes and serious errors: + [gtk] Fix several crashes, including one at start-up, that were caused by some unusual package states. - Minor bugs: + [cmdline] "aptitude why-not" no longer crashes when the final argument is a virtual package. + [gtk] The "Download started" row in the Downloads page is no longer shown. + [gtk] The dependency chains tab no longer misses some chains that "aptitude why" would show. + [gtk] Fix buffer problems due to sprintf by using aptitude's safer variant of that function. (Closes: #511559) + [gtk] When extracting just the part of a package's changelog that corresponds to versions that are newer than the currently installed version, truncate the changelog as soon as we see the version numbers go "backwards" -- that is, as soon as the previous version in the changelog is "newer" than the current version. This is important because some packages changed their version scheme over time. For instance, g++-4.2 lost its epoch in 2004 when the package name changed. But that doesn't mean that those earlier versions of the package should be displayed when we want to display the log for today's upgrade! - Cosmetic and UI bugs: + [all] The --help output now mentions --gui and --no-gui. + [gtk] Many tweaks and improvements to the appearance of the UI; particular thanks to Luca Bruno for his patches. The program looks a lot "nicer" overall as a result of his work. + [gtk] The RC style "tiny-button-style" is used for tab close buttons. + [gtk] Use the "yes" icon (a green sphere) to indicate that a package is installed and OK. This is far more recognizable for me (dburrows) than the hard disk icon we were using before. + [gtk] Not-yet-implemented menu options produce a "not implemented" message. + [gtk] Some menu items are disabled if they don't apply to the "currently selected object". + [gtk] When the user performs an action on several packages at once, any packages they install will be marked as manually installed, and package states won't change to fulfill dependencies if those dependencies are already satisfied by the selected actions. For instance, suppose that the user selects A, B, and C for installation. A depends on "D | B", and B depends on C. Previously aptitude might decide to install D to fulfill A's dependency, and would mark C as automatically installed because of B's dependency. Now, just the three selected packages will be installed, and they will all be marked as manually instsalled. - Documentation: + [doc] Wrote a new section of the documentation describing the various dependency resolution mechanisms in aptitude. This includes the old section on resolving dependencies, but also describes the immediate resolver and how to configure the full resolver using resolver hints. - Internal changes: + [gtk] Redesigned the dpkg terminal creation code to make it cleaner and more maintainable. + [gtk] Redesigned the code for the buttons in the package description pane. + [gtk] Stopped using random bits of the glade file as templates for other GUI elements, in preparation for making each tab a separate top-level widget in the file. - Translation fixes: + Fix how ngettext is used so that the plurals can be properly translated. (Closes: #505675) + Slovak (Closes: #505676) [2008-11-10] Version 0.5.0 "Wheee!" This version introduces the GTK+ frontend. Thanks are due to Obey Arthur Liu for providing code, ideas and energy to the project, and to Google for funding his work through their Summer of Code program in 2008. This is an EXPERIMENTAL, DEVELOPMENT release. It can be used for package management, but there may be bugs, there are probably places that need improvement, and it is certainly incomplete. - New features: + Integrated the GTK+ frontend written by Obey Arthur Liu for the 2008 Google Summer of Code. To manually suppress the GUI, run "aptitude --no-gui" or set the option "Aptitude::Start-Gui" to false. + Added Xapian support. Unadorned strings in search patterns now search the apt Xapian database built by Enrico Zini. There is also a new search term ?term that does the same thing. + Added the command-line option "--show-resolver-actions" to display why aptitude made the decisions it did in "aptitude safe-upgrade". - Known Regressions: + Incremental searching in the curses frontend is made much less useful by introducing Xapian. It still works, but because Xapian searches don't find substrings, the search will fail to match anything until you finish typing the entire search string. + Documentation for the GTK+ frontend has not yet been written. [2008-11-19] Version 0.4.11.11 "And the moon be still as bright" This is purely a translation release for Debian lenny. - Translation fixes: + Catalan (Closes: #499464) + German (Closes: #500444) + Greek (Closes: #498583) + Italian + Romanian (Closes: #502413) + Slovak (Closes: #498910) + Spanish (Closes: #501096, #502695, #502696, #502697) + Traditional Chinese [2008-09-05] Version 0.4.11.10 "Upon the empty winds of time" - Bug fixes: + Serious bugs: * Never, ever remove an Essential package from the curses UI without asking the user first. I chose a very simple implementation for this patch because of the pending release of lenny, and so users who are removing Essential packages will get prompted twice. But I feel this is better than occasionally not prompting at all; it will be cleanup up later, once lenny is out. + Minor bugs: * Fix displaying the section descriptions in non-UTF-8 locales. (Closes: #483464) * Add the ?task search term, which was documented in some areas but not others, and wasn't actually implemented due to an oversight. - Documentation bugs: + A whole pile of minor improvements suggested by "jidanni". (Closes: #497374, #497340, #496726, #471347, #496719, #496729, #496732, #497333, #497336, #497726, #497969) + Give the right name for the preview limit configuration option: it's Aptitude::UI::Preview-Limit, not Aptitude::Preview-Limit. + Generate the HTML documentation in a UTF-8 encoding. - Translation fixes: + Brazilian Portuguese (Closes: #496613) + Czech (Closes: #497287) + Dutch (Closes: #497965) + Kurdish + Japanese (Closes: #494816) + Lithuanian (Closes: #496504) + Norwegian Bokmål + Simplified Chinese (Closes: #497550) + Ukranian [2008-08-03] Version 0.4.11.9 "This message brought to you courtesy of the Friends of Poland society." - Bug fixes: + Minor bugs: * Don't annoy every Polish user with a warning about a badly formed string in the aptitude-defaults configuration file. Also added some documentation for translators telling them about the pitfall that led to this bug happening. (Closes: #483459) * Fix some case fallthroughs that would cause the wrong information to appear in the "archive" (%t) column for virtual packages. * Correctly handle install-and-mark-auto commands that are targeted at a particular version, like "install foo/testing+M". Previously aptitude would completely ignore the archive in this case. - Translation fixes: + German + Slovak + Swedish (Closes: #490782, #490818) [2008-07-04] Version 0.4.11.8 "Happy Fireworks Day." - Bug fixes: + Minor bugs: * Fix a long-standing and annoying bug that would cause aptitude to sometimes delete package lists if downloading new copies failed. (Closes: #201842, #479620) [2008-06-28] Version 0.4.11.7 "Yes I can" - Internal changes: * Fix several compilation errors on exotic architectures. (Closes: #488132) - Translation updates: * Basque * Portuguese (Closes: #482094) * Thai [2008-06-21] Version 0.4.11.6 "Take two" - Internal changes: * Fixed several places where the code was a bit sloppy in ways that were harmless at the moment but were turned up by -Wall -Werror. [2008-06-21] Version 0.4.11.5 "A tisket, a tasket" - Bug fixes: + Crashes and serious errors: * aptitude will no longer suggest removing Essential packages to fulfill dependencies unless you explicitly allow it to. Also, removing apt (if it is allowed) will be scored as if apt were an Essential package, meaning that it should show up last in any list of solutions. (Closes: #486748) + Minor bugs: * "aptitude add-user-tag" and "aptitude remove-user-tag" will return 0 instead of a random value when they succeed. + Translation updates: * Dutch (Closes: #486858) * Romanian (Closes: #486934) - Internal changes: * Consistently build with -Wall -Werror (it wasn't being used in all subdirectories of the code). * Clean up various minor and potential bugs turned up by increasing the warning level. [2008-06-07] Version 0.4.11.4 "Turn down the suck." - New features: * "aptitude download" now properly handles both patterns and things like version specifiers that contain a tilde. * New option "--disable-columns" to "search" that prevents aptitude from trying to reformat its output into columns. This should be useful for people trying to call aptitude from scripts. (Closes: #136874) - Crashes and serious errors: * Fix a problem with assigning scores in the dependency resolver that severely distorted the relative weights of packages. If a package conflicted with a virtual package that it also provided and replaced, aptitude would randomly score some of its versions as if they were full replacements of the current version. This could lead to surprising and wrong resolver outputs. (Closes: #483920) - Documentation bugs: * Fix some broken internal links. * The manpage formatting of elements is correct now. They're bolded, and literal file extensions don't get misinterpreted as groff escapes. (Closes: #473580) - Translation updates: * Install defaults files for locales that contain an underscore, like pt_BR. (Closes: #483620) * Russian. (Closes: #483943) [2008-05-26] Version 0.4.11.3 "Not looking back." - Crashes and serious errors: * "unhold" should work now. (Closes: #477165) * Fixed viewing changelogs at the command-line, which was almost totally broken in recent releases. (Closes: #481458) * Fix getting the changelogs of bin-nmued packages from the curses interface. (Closes: #333468) For some reason when this was fixed for the command-line mode, the equivalent change didn't make it into the curses codebase. * Don't die with an assertion error when --show-why is used in the presence of impure virtual packages. - Cosmetic and UI bugs: * Always be at least as quiet as the user requested; when aptitude automatically enabled quietness if its output wasn't a TTY, it could actually become less quiet than otherwise! (Closes: #476749) * If the user asks for the justification of a manually installed package, try to find a nontrivial answer (some other package that requires it) instead of just telling them that it's already installed. (Closes: #477038) * Remove an incorrect hyphen in the output of --help. (Closes: #476835) * When the resolver is run several times in a row without user interaction (for instance, in "safe-upgrade"), only print "Resolving dependencies..." once. * When listing the complete chains of dependencies that are holding a package on the system, show "A provided by B" as "A <-P B", not "AP<- B". (i.e., add a space after "A") - Build system fixes: * VPATH should be supported better in doc/. * Most files will generate Doxygen output now (they were missing \file tags). - Documentation bugs: * Clarify how aptitude's search language behaves in some syntactic corner cases, like "~nname?installed": the "?" is part of the string parameter to the first matcher, not the start of a new match term "?installed". * Change the documentation of "aptitude why": it shows why packages should be installed, not why they can be installed, and be clearer about what happens when "why" is called with a single argument. * Fix the manpage to talk about Recommends-Important instead of Install-Recommends. (Closes: #480533) - Translation updates: * Brazilian Portuguese (Closes: #481007) * Danish (Closes: #476732) * French * Galician (Closes: #476837) * German (Closes: #476344) * Japanese * Norwegian Bokmal (Closes: #480063) * Polish (Closes: #480062) * Simplified Chinese (Closes: #475740) * Vietnamese (Closes: #477295) [2008-04-09] Version 0.4.11.2 "How far the sky, how cold the night, how still the flowing river. How sharp the air -- so pure the light -- beneath Orion's quiver. Beneath the frozen, staring, starry sky we lay a-shiver." - New features: * The information area can display tabs for the alternative views it supports. This is off by default because I found that it was too intrusive (mainly because it appears between the short and the long description of a package). If cwidget gets the ability to put tabs at the bottom of multiplex widgets, I might change this. - Documentation bugs: * Fix several XML errors in the manpage source (Closes: #473722) - Bug fixes: + Crashes and serious errors: * Make the safe resolver not explode exponentially when new Recommendations are present. The root cause here was that the "mandate" mechanism the safe resolver uses to ensure that it progresses monotonically towards a solution doesn't work in the presence of Recommends, so the resolver ended up exploring the entire space of possible resolutions to packages' Recommendations. (Closes: #474680) * Don't crash if a package's Section is empty. (Closes: #474115) + Cosmetic and UI bugs: * The "why installed" information display in the package list will now display an informative message when no package is selected, the same way that the "related dependencies" display does. * The "why installed" information display should work better for newly installed packages now. * "aptitude --version" prints more information about the libraries it was compiled against. * "Help -> About" prints a correct copyright date now. * The status indicator has been rewritten so that it no longer needs numeric conjugation (meaning it won't say "there are 1 update"). (Closes: #486186 and friends) * "why" and "why-not" now appear in the list of commands in "aptitude --help". (Closes: #454088) + Minor bugs: * Use a 'latch' configuration option to migrate from Recommends-Important, instead of just clobbering the old option. (Closes: #473872) * Treat packages that were removed but whose configuration files remain on the system as if they're not installed in the output of --show-why. This fixes some cases where --show-why would hide the reasons for some installations. - Translation updates: * Galician (Closes: #474672) * Vietnamese (Closes: #473719) [2008-03-30] Version 0.4.11.1 "No, no, fool, I said bring me the FLUFFY Bunny Slippers of EXTREME Dismemberment, not the EXTREME Bunny Slippers of FLUFFY dismemberment! Does an evil overlord have to do everything himself around here?" - New features: * The command-line argument --show-why will cause all installation and removal commands to display a brief summary of the dependencies related to an installation or removal. For instance: The following NEW packages will be installed: libboost-iostreams1.34.1{a} (for wesnoth) wesnoth wesnoth-data{a} (for wesnoth) wesnoth-music{a} (for wesnoth) Note that wesnoth-music is a dependency of wesnoth-data, not wesnoth. --show-why displays the manually installed package behind each automatic installation. If -v is passed on the command-line, it displays the entire chain of dependencies leading to each manual package. --show-why also handles upgrades: The following packages will be upgraded: klibc-utils libklibc (for klibc-utils) --show-why is limited to the capabilities of the logic behind "aptitude why". Although libklibc might be the package that was originally marked for upgrade, --show-why has no knowledge of this fact: it just knows that klibc-utils requires libklibc. * In the ncurses interface, added a menu entry to cycle the display in the lower pane (equivalent to pressing "i" but more discoverable). * The ncurses interface now displays the number of packages in a tree next to the tree header, and in the information area when the header is highlighted. * The header for the list of versions in the ncurses interface now reads "Versions of " instead of just "Versions". * A new extract-cache-subset command that will create a reduced copy of the package cache, removing all but a given list of packages and all references to packages not in the list. It's intended for, e.g., generating test cases for package managers. - Bug fixes: + Crashes and serious errors: * Handle exceptions thrown by the Debtags constructor (e.g., when debtags has been purged and there are no data files) instead of just crashing. (Closes: #472695) * Eliminate a case where we would access uninitialized memory while starting up, found thanks to valgrind. + Cosmetic and UI bugs: * Correct the documentation within the on-line help of how to leave the on-line help (it was never updated when the help went from being a dialog to being a top-level view). * Correctly save changes to string configuration options. (Closes: #471315) * Don't print an error on startup when debtags isn't installed. (Closes: #472678) * Use less technical language when "why" fails to find a derivation for "A transitively requires B". * Eliminated some cases where description signals for tree headings weren't being connected. These weren't noticable in the past because in most cases, the heading had no description anyway. - Translation updates: * The aptitude-defaults.* files are now installed to /usr/share/aptitude. (Closes: #472625) * Simplified Chinese (Closes: #458162, #473363). * Vietnamese (Closes: #473229) [2008-03-15] Version 0.4.11 "When you're tired of being beat with a stick, you're tired of Debian." - New features: * Search terms can be named with words instead of single-character flags. Each new-style search expression starts with a question mark ("?"), followed by the name of the matcher; for instance, "?obsolete" is equivalent to the old-style expression "~o". At the command-line, package names containing question marks are treated as search expressions, just like package names containing tildes ("~") are. The old-style syntax is still present and works just like it did previously. See the reference manual for complete documentation of the new syntax. This will hopefully make the search syntax more memorable and make it clearer what a given search expression actually does. In addition, this opens up the possibility of greatly expanding the number of search patterns supported by aptitude (the requirement to choose a meaningful and unique single character had become a major limiting factor in the ability to add new search terms). * New search terms: + ?source-package(expr) matches packages whose source package matches the given regular expression. + ?source-version(expr) matches packages whose source version matches the given regular expression. + ?all-versions(expr) matches a package if expr matches all versions of that package. + ?any-version(expr) matches a package if expr matches any single version of that package. + ?user-tag(pattern) matches a package if it has an attached user-tag matching the given regexp (see below). + ?for var: expr binds var inside expr to the package or version being tested. + ?bind(var, expr) matches anything if the package or version bound to var matches expr. + ?=var matches the package or version bound to var by an enclosing ?for. For instance, "?for x: ?depends(?recommends(?=x))" will match any package X that depends on a package that recommends X. * aptitude now supports attaching arbitrary strings to packages, known as "user tags" (to distinguish them from debtags tags). The following commands will manipulate user tags: aptitude add-user-tag tag package... aptitude remove-user-tag tag package... Each of these commands will add user tags to or remove them from one or more packages (possibly selected using search expressions). In addition, all command-line actions that modify package state now take the following optional arguments: --add-user-tag TAG --add-user-tag-to TAG PATTERN --remove-user-tag TAG --remove-user-tag-from TAG PATTERN The variants that take a PATTERN will add tags to or remove tags from any packages that match the given PATTERN. (e.g.: "--add-user-tag-to installed-for-build-dep ?action(install)"). The variants that do not take a PATTERN will affect any package that is being modified (this is equivalent to using the pattern "?not(?action(keep))"). The ?user-tag(tag) matcher will select packages with a tag matching the given regular expression. User tags also show up at the end of package descriptions, next to the list of debtags tags for a package. * All command-line actions that modify package state now accept the arguments --safe-resolver and --full-resolver. --safe-resolver forces the command to use the same resolver logic as --safe-upgrade (e.g., "aptitude install --safe-resolver exim4" to install exim4 as long as it can be installed without removing any packages). --safe-resolver can be enabled in the config file by setting Aptitude::Always-Use-Safe-Resolver to "true"; passing --full-resolver will override this configuration option. The option --no-new-installs and the new option --no-new-upgrades will control whether the safe resolver attempts to install new packages or upgrade installed packages. Future versions of aptitude will enable this logic in the curses UI as well, but I haven't decided how it should be exposed yet. * The aptitude dependency resolver will now refuse to adjust held packages or install forbidden versions unless you manually allow it to. This behavior can be disabled by setting Aptitude::ProblemResolver::Allow-Break-Holds to "false". aptitude will still break holds when packages are being automatically installed; there is a pending patch against apt that eliminates this behavior. * The aptitude dependency resolver will add a bonus to solutions that remove a package and install another package that fully replaces it by declaring a conflicts/provides/replaces relationship. (Closes: #466374) * aptitude now has a "build-dep" command that will install build-dependencies from the command-line. (Closes: #243317) * A new "subdirs" variant of the section grouping policy is introduced in this release, courtesy of Paul Donahue, and is enabled by default. This will behave just like the old sectioning policy, except that if a section has several components (for instance, games/arcade/space), they will all be realized as tree levels in aptitude. A side effect of this change is that section descriptions are no longer hard-coded in the source (see the documentation of Aptitude::Sections::Descriptions). * Command-line updates in aptitude will now list packages that are newly obsolete. This doesn't work when a source is removed and all its packages become obsolete, for technical reasons. * aptitude now uses libept to handle debtags information instead of a slow and wrong internal implementation. (Closes: #397652, #406201) "aptitude update" will merge new package information into the debtags database automatically by running "debtags update --local". - Bug fixes: + Cosmetic and UI bugs: * The curses interface now uses a spinner to indicate download progress, since the percentage measurement can't be made correct in the current apt model. * Remove some stray cw:: strings that made it into user-visible messages (thanks to Jacobo Tarrio for pointing them out). * Interpret %-escapes in the descriptions of configuration options instead of displaying them to the user; thanks to Jens Seidel for pointing this out. * At the command-line prompt, don't list packages that aren't being upgraded unless the user tried to upgrade them. e.g., "aptitude install foo" will no longer spew the entire list of pending upgrades to the terminal. * When displaying dependencies or version numbers at the command-line (i.e., -D or -V is passed), add an extra space between packages, to make it a little more obvious which tags "belong" to which package. * If the resolver is allowed to produce the solution 'cancel all actions' (i.e., "Aptitude::ProblemResolver::Discard-Null-Solution" is false), the brief indicator in the UI will say that it cancels all the pending actions instead of counting its component actions. * When displaying the packages that could satisfy a versioned dependency, don't list packages that provide the package name unless they declare a versioned provides that matches the dependency's version restriction. (Closes: #464131) * Don't try to generate and then parse a matcher when searching for packages that look like what the user typed; instead, write explicit code to compare against package names and descriptions. This allows aptitude to find similar package names even if the corresponding command-line argument is an invalid search pattern (without having to write complex escaping logic); e.g., "aptitude install +5" does something sensible instead of printing a confusing error message. * If the user enters "q" during command-line dependency resolution, quit the program instead of falling back to manual resolution; manual resolution is only used if automatic resolution hits a fatal error or if the user requests it by typing "x". (Closes: #459629) + Translation bugs: * Generate POTFILES.in automatically, so it stays up-to-date without requiring manual intervention. * Flag W_() as a translation marker; failure to do this caused a lot of strings to be incorrectly left out of the translation files; pointed out by Jens Seidel. * Fix translation of strings displayed by the options editor (they weren't getting translated at all); thanks to Jens Seidel for reporting this. + Minor bugs: * Fix task handling in the case that different versions of a package are in different tasks. (Closes: #459348) * "safe-upgrade" will now attempt to automatically maximize the upgrade. Hopefully this will eliminate the situation where you run "aptitude safe-upgrade" and then discover that there are still some upgradable packages. + Crashes and serious errors: * aptitude will now build with g++ 4.3, assuming that 4.3-compatible versions of cwidget and sigc++ are installed. (Closes: #452204, #452540) * Reduce the translation percent threshold for the various versions of the manual until it compiles. (Closes: #470054) * aptitude now uses Apt::Install-Recommends instead of Aptitude::Recommends-Important to control whether recommendations are automatically installed. (Closes: #458189, #448561) Old configurations will be migrated to the new configuration option, if possible. Among other things, this means that --without-recommends works again (it was broken when recommends handling moved to apt). * Ensure the resolver state is always synchronized with the package cache, eliminating some cases where the resolver would break or produce wrong answers. (Closes: #421395, #432411) * Don't crash at the command-line when displaying the version numbers of packages being removed that don't have an installation candidate; also, the version display for removed packages is now meaningful. (Closes: #459336) * Don't crash at the command-line when displaying the version numbers of packages in the ConfigFiles state that are being purged; instead, display "Config Files" as the removed version. (Closes: #461669) As a side note: the code in question has been adjusted to be robust against bad version pointers, and should display "??" instead. If you see this, it's a bug, but at least aptitude won't crash in this case. * Don't crash in "aptitude update" when the package lists can't be parsed for some reason; instead just go ahead and download new ones. (Closes: #468751) + Documentation bugs: * Fix the documented default keybinding for PrevPage. * The documentation of searching has been substantially rewritten, and is hopefully much more useful now. In particular, there's more information about how versions are handled, and a concise table listing all the search expressions supported by aptitude. * Document some of the files aptitude uses in a FILES section in the manpage (Closes: #470839). + Translation updates: * Basque * French - Christian Perrier: Fix the use of quote in the French translation (Closes: #460808). * Galician * German (thanks to Jens Seidel) * Japanese (thanks to Noritada Kobayashi) * Norwegian Bokmål * Portuguese * Romanian (thanks to Eddy Petrișor) * Russian * Simplified Chinese * Slovak * Spanish * Vietnamese [2007-12-15] Version 0.4.10 "Oscillating Reindeer" - New features: * In command-line mode, if the resolver fails to produce a solution for whatever reason, then instead of aborting the program entirely aptitude will now display a prompt at which you can fix the dependency problems by hand. Enter "r" at this prompt to try to automatically fix dependencies again. * safe-upgrade will now install new packages to fulfill dependences (but it will never remove packages, downgrade packages, or install a version that's not the default). The option --no-new-installs will disable this behavior. * Updates and commands that install, remove, or upgrade packages will now display a brief summary of what changed. For instance: There are now 64 updates [+10], 3 new [+1]. Unfortunately, this change requires reading the cache after an update is complete. Passing -q will disable this behavior, but also make the update progress bar less attractive. More work on resolving this tension between features and performance remains to be done. * Add an option --allow-untrusted to override trust warnings. (Closes: #452201, #452541) * Recommended packages are now hidden if quiet mode is enabled. (Closes: #452202) * The options Aptitude::ProblemResolver::Trace-File and Aptitude::ProblemResolver::Trace-Directory allow you to generate a minimal cut of the cache that allows a problem resolver run to be reproduced. The eventual aim is both to simplify bug reporting and to generate a corpus of automatic test cases for the resolver (although more work needs to be done to accomplish the latter). * When run in command-line mode, instead of displaying many separate and sometimes duplicative groups of packages (e.g.: installed, auto-installed, auto-installed in a light cream sauce, etc), aptitude displays packages in a few exclusive categories and uses tags (like the existing "purge" tag) to provide more state information. For instance: $ aptitude -s install wesnoth The following NEW packages will be installed: wesnoth wesnoth-data{a} wesnoth-music{a} $ aptitude -s remove freeciv-data (... dependency resolution ...) The following packages will be REMOVED: freeciv-client-gtk{a} freeciv-data freeciv-server{a} ggzcore-bin{u} libggz-gtk1{u} libggz2{u} libggzcore9{u} libggzdmod6{u} libggzmod4{u} Here {a} indicates that a package was automatically installed or removed, and {u} indicates that a package is being removed because it is unused. Hopefully this will be less confusing than the old format. - Bug fixes: + Crashes and serious errors: * Track down and fix a SEGV triggered on the first action after a cache reload (sometimes). There are multiple bugs where this might have been the root cause, but it's confirmed that it was the cause of at least #454695, #454700, #455349, and #453362. Bug #455865 is almost certainly the same issue, and bug #352278 may be another manifestation of it. * If StepLimit is set to 0, refuse to solve dependencies instead of going into an infinite loop. (Closes: #451311) + Minor bugs: * Return a failing exit code after jumping from the command-line to visual mode if the last install run failed. (Closes: #282408) * Return a failing exit code from "aptitude update" if any download fails. (Closes: #233129) + Cosmetic and UI bugs: * In addition to Enter, Space will now activate checkboxes and radio buttons. (Closes: #451765) * Abort the program if we get EOF at the resolver prompt instead of claiming the resolver failed. * Command-line searches will now only print each result once. (Closes: #450798) + Documentation bugs: * The options menu documentation now minimally describes the new configuration interface. * The documentation of the configuration file options has been placed back in alphabetic order after apparently suffering bitrot over the years. * Corrected the documented default of Parse-Description-Bullets to match reality. * The manpages for aptitude-create-state-bundle and aptitude-run-state-bundle are now generated using DocBook, which should produce higher-quality and more consistent typesetting, as well as making the manpages more maintainable and allowing translators to easily produce localized versions of the manpages. - Internal changes: * Removed some unit tests that really tested cwidget. - Translation updates: * French (Christian Perrier) [2007-11-17] Version 0.4.9 "Ow. Ow. Ow. Ow. Ow." - New features: * Add a matcher ~o for obsolete/local packages. (Closes: #397547) - Bug fixes: * Correctly set the name, description, and long description of radio options. (Closes: #449138) * Don't remove *.gmo in distclean. Apparently this causes trouble for translations. (Closes: #451584, #441696) [2007-11-15] Version 0.4.8 "The fun never stops!" This change removes the internal widget set, instead using the cwidget curses widget library (which just happens to provide all the same functionality as the aptitude widget set, imagine that). - Bug fixes: + Cosmetic and UI bugs: * Interpret key names in the broken indicator as wide-character strings; fixes actual display corruption on some platforms and potential corruption on all platforms. (Closes: #448753) [2007-10-30] Version 0.4.7 "Where did all these balloons come from? And why am I wearing a fake nose?" - New features: * The options dialogs have been completely replaced by a new interface, based on a top-level list view. This fixes many deficiencies of the old interface: it handles long strings more gracefully, avoids many of the focus-handling bugs that the old dialogs had, and should generally be better-behaved. (Closes: #197976, #331200, #424708) * Prompts that ask you to enter text will now wrap to multiple lines when the text gets long, rather than hiding parts of the string. * The online help and other Help-menu items are now top-level views, which should make them somewhat more usable. (Closes: #434349) * Support for the "Breaks" field and for trigger states (thanks to Michael Vogt and Ian Jackson for patches and prodding). (Closes: #438548) * Two new styles, "PkgDowngraded" and "PkgDowngradedHighlighted", are provided to control how downgraded packages look. By default these packages look like any other installed package. (Closes: #439924) * aptitude can now display homepage URLs stored in the Homepage field of packages. This requires a recent version of apt; 0.7.8 or better includes it. - Bug fixes: + Crashes and serious errors: * When applying the resolver's output, only set the packages that are newly installed to be automatic, rather than making everything the resolver touches automatic. * Save and restore the automatic flag on packages that are not not currently installed and that are being installed. (Closes: #435079) * Fix the help generated when the user presses '?' at the prompt to not segfault, and rewrite the code to avoid this bug in the future (it was using a format string with a huge number of placeholders; now it builds a list of the output lines explicitly). * Eliminate a crash on startup caused by using a global pointer to the cache in the package matching logic (which might be null) instead of taking a valid pointer to the cache as a parameter. This is necessary since the matcher might be invoked while the cache is being loaded, e.g., to decide if a package should be part of the root set. + Minor bugs: * Fix the 'pattern' grouping policy: it was inserting packages into all the trees that matched, not just the first one. * Make removals and holds that occur later on the command-line override earlier commands. * The debugging output from "why" no longer tries to dereference invalid pointers or display multibyte strings as single-byte strings. (Closes: #447290) + Cosmetic and UI bugs: * If the text in a column of the "why" output exceeds the column width, the column will now wrap correctly onto the next line. * Auto-held and unconfigured packages no longer generate ?????? in the aptitude log. + Documentation bugs: * Fix the documentation on aptitude development. It now gives the correct URL and version control system for the upstream repository, and somewhat more actively solicits contributions. - Translation updates: * Galician (Closes: #446620) * Nepali * Russian * Spanish * Swedish (Closes: #434643) [2007-07-25] Version 0.4.6.1 - Minor bugfix to include the new scripts/manpages that didn't get distributed with the previous release. [2007-07-25] Version 0.4.6 "He who works and works all day, gets to work another day." - New features: * Added two programs, aptitude-create-state-bundle(1) and aptitude-run-state-bundle(1), to eliminate some of the drudgery involved in collecting apt state and running aptitude with a state snapshot. * "why" output is now available in visual mode (hit 'i' to cycle through informational displays). The formatting and styling are pretty awful in this release. - Bug fixes: + Crashes and serious errors: * Handle EINTR from select() better; thanks to Jiří Paleček for tracking this down. (Closes: #431054, #431688, #432323) Note that there is still a race condition lurking here, although it's almost impossible to trigger it. The next release should eliminate it, though. * Run mark-and-sweep on startup, to initialize apt's auto flags. This should prevent aptitude from losing the auto flag when a package is marked for upgrade. (Closes: #432017) * Don't crash on startup when something goes wrong early in the initialization process (e.g., when the cache is locked; Closes: #430061). + Minor bugs: * Disable unused-package removal if Delete-Unused is false. (Closes: #431716) + Cosmetic and UI bugs: * Don't garble descriptions in non-UTF8 locales. (Closes: #432911) * Hopefully fixed the problem that was garbling the startup progress indicator. * Eliminate another bogus error about a supposedly locked cache. (Closes: #431909) + Documentation bugs: * Fix the manpage's metainformation so that the footer gets generated. + Potential bugs: * Handle some unusual cases where the resolver could dereference invalid pointers (no bugs reported). * Added code to recover with an error if apt improperly marks a dependency as broken. - Translation updates: * Basque (Closes: #432535) * Dzongkha * French * Vietnamese (Closes: #432283) [2007-07-03] Version 0.4.5.4 "Oh my God, THEY KILLED LENNY! (actually, it was sid, but I wanted to say that)" - New features: * Expose the "why" command from the command-line Y/n prompt. Typing "w " will invoke "why ". The resolver prompt doesn't support this, maybe it should? - Bug fixes: + Crashes and serious errors: * Eliminate a bounds error that was crashing the vs_editline. (Closes: #429673) + Minor bugs: * Force aptitude to write out a state cache the first time it runs. Without this change, it won't know which packages are "new" until after the first time the user installs something. (Closes: #429732) + Build system fixes: * Add a configure check for the new apt, so users get more sensible build errors with incompatible apts. * Have 'make clean' remove autogenerated XML files, so they don't show up in the Debian diff. - Translation updates: * Basque (Closes: #418862) * Galician (Closes: #429504) * Vietnamese (Closes: #429447) [2007-06-17] Version 0.4.5.3 "The Long Dark Tea-Time Of The Code" - New features: * Added two commands "aptitude why" and "aptitude why-not" that provide explanations of why a package is, must be, should be, or must not be installed on your system. They are not complete, by which I mean that they can't always provide the most useful justification, but they should answer a lot of questions that I hear people asking regularly. - Bug fixes: + Crashes and serious errors: * Eliminate a nasty race that was probably the cause of corrupted output on some dual-core systems. (Closes: #414838, #406193) * Fix a bug that could cause aptitude to try to remove a package that it had already removed. (Closes: #429388) This can't be totally eliminated without help from dpkg, but hopefully this fix will make the problem much more difficult to trigger. Unfortunately, the fix also means that aptitude has to perform a save/load cycle after installing packages to make sure that everything is consistent. * Allow "keep" actions to be undone. * Don't consider every removed package to be an unused package. * Clear the broken package indicator when the user undoes an action. * Fix build errors on other architectures. (Closes: #429348) * Set packages to manual mode when the user cancels their removal. (Closes: #429271) * Fix compile bugs with g++-4.3. (Closes: #413488) [2007-06-15] Version 0.4.5.2 "To ook is human, to meow feline" - Bug fixes: * Multiple minor bugs introduced by the patches to support apt's auto-marking were fixed. - Packages kept at their current version by applying a solution no longer get their hold flags turned on. - Packages that are installed by a solution get their auto-installed flag set. - Applying a solution updates the resolver state (so the error bar doesn't stick around until you change a package's state by hand). * Fixed a broken cross-reference in the manpage. [2007-06-15] Version 0.4.5.1 "Nothing could possibly go wrAAAAAAAAAAaaaa..." - New features: * aptitude now uses apt's central database to track which packages are auto-installed. Aptitude's own list of automatic packages will be merged into the global list the first time that aptitude is run. * Packages that are not fully configured are now listed in aptitude's preview, and aptitude will let you do an install run if there are unconfigured packages but nothing to install/remove. (Closes: #424709) - Translation updates: * French (Closes: #428585, #428826) * Portuguese (Closes: #425779) * Vietnamese (Closes: #426976) [2007-05-14] Version 0.4.5 "Confused Cockroach" - New features: * For standard string prompts, the default is now initially displayed, but erased if the user starts out by inserting a new character. * 'N' now repeats the last search in the opposite direction (as opposed to 'n', which repeats it in the same direction). (Closes: #414020, #397880) * aptitude now recognizes Apt::Get::AllowUnauthenticated as a synonym for Aptitude::Ignore-Trust-Violations. (Closes: #411927) * Fix the handling of dselect/dpkg state. Previously, mixing the command-line and visual modes of operation could result in aptitude thinking it should remove a package that was actually installed on your system. This was especially nasty since the Debian installer tripped this condition, leaving people with a broken initial system. Ow. (Closes: #411123) * Added a config option "Aptitude::CmdLine::Resolver-Show-Steps", which causes the dependency resolver to default to displaying its output as a list of individual resolutions, rather than a list organized by the type of change that will be made. (equivalent to pressing "o" at the first resolver prompt) * Added "safe-upgrade" and "full-upgrade" as synonyms for "upgrade" and "dist-upgrade" respectively. The goal here is to eliminate massive confusion about what these commands actually do. "upgrade", in particular, is now deprecated ("dist-upgrade" probably has too much currency to kill off, but hopefully people will at least be rid of the notion that it's exclusively for whole-distribution upgrades). * The option "Aptitude::Get-Root-Command" can be used to choose how aptitude tries to gain root privileges. It defaults to "su"; setting it to "sudo" will choose that program, and other programs can also be chosen (see the user's manual for details). - Bug fixes: + Crashes and serious errors: * Improve the internal consistency of aptitude's reduction of the apt dependency graph. As part of this change, a new (undocumented, primarily for debug purposes) command-line action "check-resolver" was added; it verifies empirically that the dependency structure obeys the invariants that it's expected to obey (for instance, that all forward dependencies have a corresponding reverse dependency). In addition to a number of harmless discrepancies, this flushed out the forward/reverse dependency mismatch that was causing crashes in the resolver. (Closes: #420358, #420381, #420407). Note that check-resolver only verifies *internal* properties of the resolver's model; it doesn't verify that the model faithfully represents the dependency problem posed by the apt package database. * Don't crash on amd64 and other architectures where va_args structures can't be reused. This is an old bug; it was turned up by a new test for ssprintf. * Increase the default value of StepScore to 70 and improve its documentation. This should make it much less likely that the resolver wanders off into lala land trying to solve dependency problems. (Closes: #418385) * When --without-recommends is passed at the command-line, automatically enable Keep-Recommends at the same time we disable Recommends-Important, so we don't automatically remove half the user's system. (Closes: #143532) * Fix an internal sprintf variant to produce correct output (the previous code was safe, just wrong; it safely produced a buffer of the correct size, then truncated it to its initial buffer size). * If the dependency resolver throws an exception, catch it and handle it as cleanly as possible, rather than unceremoniously exploding. * Don't block signals when running dpkg after a package install failed, and tell dpkg not to use TSTP in this circumstance. This tended to cause horrible breakage of dpkg's auto-backgrounding (the "Z" conffile prompt option). (Closes: #367052) * Fix a crash that occurred when generating a revdep-count column for some virtual packages. (Closes: #420405) * Don't abort() if something tries to display a progress percentage that's not strictly between 0 and 100. (Closes: #425145) + Cosmetic and UI bugs: * If a package is both "new" and "upgradable", place it in the "upgradable packages" list instead of the "new packages list", since this seems to be the path of least confusion. (Closes: #419999) * Try to include more information when an internal consistency check is triggered (e.g., the dependency that the program was looking at). * Only print an error about being unable to acquire a lock after downloading when an error actually occurred during the call to GetLock(); this message was getting triggered by stray errors from methods invoked earlier in the download/install process. (Closes: #422700) * Don't ever display more than one "really quit?" dialog box, so people who hold down the "q" key don't have to also hold down "n". (Closes: #411158) * When writing out the aptitude state file, open it mode 0644 and explicitly chmod it to mode 0644, so it doesn't start out world-writable, but ends up group- and world-readable even if root has a more restrictive imask. pkgstates isn't security-sensitive, and aptitude produces surprising results if it can't read this file. If you really need to hide pkgstates from non-root users for some reason, you can chmod -x the directory /var/lib/aptitude. (Closes: #41809, #421811) * If a command (e.g., update) takes no arguments, aptitude will produce an error if it receives arguments. (Closes: #411152) * Write "[Enter]" instead of "enter" in the media-change prompt, to make it clear that the user should press the key of that name. (Closes: #414777) * Remove *.gmo in distclean, so the Debian package can be built twice in a row. (Closes: #424102) + Potential bugs: * Delete some internal cache arrays using operator delete[] instead of operator delete. * Initialize package selection states to the Unknown state explicitly. + Documentation: * Fix the docbook encoding of the users manual and manpages. * Fix an error in the DocBook code of the manual page that caused the AUTHOR section to appear twice. * Fix how bolding is applied to literal elements in the manpage, so that the manpages are no longer screwed up. (Closes: #415468, #416232) * Users manual corrections (thanks to Noritada Kobayashi for patches). * Improve the description of the Recommends-Important, Keep-Recommends, and Keep-Suggests: clean up the language and try to make their precise behavior clearer. Thanks to Stefan Kangas for the initial suggestion. (Closes: #405002) * Try to make it clearer in both the manpage and the user's manual that "aptitude search" does an implicit OR on its command-line parameters, rather than an implicit AND. (arguably it SHOULD do an "AND", but it's probably too late to change this now) * Add some commands to the online help that weren't previously documented. (Closes: #402360) * Fix misaligned items in the online help. (Closes: #399048) - Translation updates: * Task group names localized. * Basque (Closes: #418862) * Catalan (Closes: #353308) * Chinese (Simplified) (Closes: #405166) * Chinese (Traditional) (Closes: #338056, #418139) * Danish (Closes: #409480) * Finnish (Closes: #392305) * French (Closes: #403561, #388504) * Galacian (Closes: #412829, #425358) * Greek * Hungarian (Closes: #405550) * Japanese users manual added. * Marathi (Closes: #416807) * Nepali (Closes: #452278) * Portuguese (Closes: #418940) * Romanian (Closes: #415763) * Slovak (Closes: #401105) * Spanish (Closes: #416339) * Ukranian (Closes: #415448) [2006-10-26] Version 0.4.4 "I shall smite you with my Ultimate Power of Cuteness!" - New features: * "unhold" now just clears the hold flag; it doesn't try to upgrade the package that was just unheld. * The grammar used to parse bulleted lists has changed. Full stops are no longer considered to be paragraph breaks or otherwise significant unless they are indented by exactly one space (i.e., unless they are significant in the standard grammar). A second change introduced by this patch is that a full stop by itself on a line no longer terminates a bullet list; the list will continue as long as the text after the paragraph break is indented at the same level as the rest of the list. This fixes some cases where aptitude badly mangled descriptions that used to work, and means I feel safe re-enabling bullets by default. (Closes: #388594) - Bug fixes: * By default, don't remove Linux kernel images that are unused. (Closes: #386307) * Generate a more useful error message for corrupted or unverifiable file downloads. (Closes: #387537) * Fix minibuffer messages so that they go away when you press a key. (Closes: #395201) * Apparently time is allowed to go backwards, so don't assert that it doesn't in the test suite. (Closes: #381481) * Document what "unhold" does. (Closes: #387336) * Consistently use instead of