~ubuntu-branches/ubuntu/maverick/aptitude/maverick

2005-10-01  Daniel Burrows  <dburrows@debian.org>

	* NEWS:

	  Add some missing entries and bug numbers.

	* Makefile.am:

	  Distribute the other READMEs.

	* NEWS:

	  Mention the build script migration.

	* configure.ac, src/aptitude.h:

	  Deal with the fact that the old gettext/autoools generated a

	  test for locale.h and setlocale but the new one doesn't.  I

	  could stop using the test, but instead I decided to just test

	  explicitly for them and define the appropriate variables if

	  they're found.

	* src/aptitude.h:

	  Instead of using #define to empty setlocale, inline it.  This

	  still fails to compile, but the error is a lot more obvious.

	* ***:

	  Rebuilt all of the autotools and gettext junk from scratch using

	  automake-1.9.  In theory this will make VPATH builds work

	  (xgettext was failing); however, I'm going to wait and see.  It

	  does mean that I am no longer using old crufty versions of the

	  tools, though.

	* Makefile.am:

	  On the other hand, the non-generated manual files really *are*

	  in $(srcdir).

	* doc/en/Makefile.am, doc/fi/Makefile.am, doc/fr/Makefile.am:

	  Also pick up README from '.', not $(srcdir).

	* doc/cs/Makefile.am, doc/de/Makefile.am, doc/en/Makefile.am, doc/fi/Makefile.am, doc/fr/Makefile.am:

	  Whoops, there is no $(builddir) -- it's just '.' in rules.

	* doc/cs/Makefile.am, doc/de/Makefile.am, doc/en/Makefile.am, doc/fi/Makefile.am, doc/fr/Makefile.am:

	  Consistently put the output directories in $(builddir), not

	  $(srcdir).

	* src/mine/Makefile.am:

	  Use $(wildcard) instead of listing all the help texts

	  explicitly.

	* Makefile.am:

	  Use $(wildcard) for some wildcarded filenames so that VPATH

	  builds work.

	* doc/fi/Makefile.am, doc/fr/Makefile.am:

	  Fix up some more documentation Makefiles that I missed

	  previously.

	* doc/cs/Makefile.am, doc/de/Makefile.am, doc/en/Makefile.am, doc/fi/Makefile.am, doc/fr/Makefile.am:

	  Fix the documentation Makefiles to refer symbolically to the

	  souce directory using $(srcdir) instead of directly via '..' and

	  '.'.

	* NEWS:

	  Add a comment about whitespace in patterns, change the quote.

	* Makefile.am:

	  Put intl before the other subdirectories, so builds with an

	  included libintl work.

2005-09-30  Daniel Burrows  <dburrows@debian.org>

	* src/download_screen.h:

	  Fix the #include of config.h. (done previously but I forgot to

	  save!  ah the irony)

	* configure.ac:

	  Bump the version to 0.3.4.

	* src/mine/Makefile.am:

	  Explicitly list all of the minesweeper help files -- apparently

	  automake's "make dist" target doesn't expand wildcards.

	* src/mine/Makefile.am:

	  Add top_builddir to Minesweeper's include path as well.

	* src/generic/util/Makefile.am:

	  Add slotarg.h to the list of sources for libgeneric-util.

	* src/aptitude.h, src/main.cc, src/pkg_ver_item.h, src/ui.cc, src/vscreen/curses++.h, src/vscreen/vscreen_widget.cc:

	  Change a bunch of includes of config.h to use <>, and protect

	  the ones that weren't with #ifdef HAVE_CONFIG_H.

	* src/generic/util/Makefile.am:

	  Add bool_accumulate.h to the list of generic sources.

	* src/vscreen/Makefile.am:

	  Add ref_ptr.h to the list of vscreen sources.

	* src/generic/problemresolver/Makefile.am:

	  Forgot to add resolver_undo.h to EXTRA_DIST so that it gets

	  distributed; do so.

	* src/Makefile.am, src/cmdline/Makefile.am, src/generic/apt/Makefile.am, src/generic/problemresolver/Makefile.am, src/generic/util/Makefile.am, src/vscreen/Makefile.am, src/vscreen/config/Makefile.am, tests/Makefile.am:

	  Include $(top_builddir) so we can get config.h in VPATH builds

	  (e.g., in 'make distcheck').

	* doc/en/aptitude.xml:

	  Correct a strange turn of phrase.

	* src/vscreen/Makefile.am:

	  Remove references to some .h files that got moved down to

	  generic/.

	* src/vscreen/vscreen_widget.cc:

	  Allow destroy() to be called multiple times but ignore it after

	  the first call.  This allows download widgets to destroy

	  themselves immediately upon being cancelled, even though the

	  download process will also issue a destroy command on its own.

	* src/ui.cc, src/ui_download_manager.cc, src/ui_download_manager.h:

	  Fix more compile problems.

	* src/download_list.cc, src/download_list.h:

	  Fix compilation errors.

	* src/download_list.cc:

	  Actually use abortslot when a download is cancelled.

	* src/ui.cc, src/ui.h, src/ui_download_manager.cc, src/ui_download_manager.h:

	  Expose the widget that's handling the download status, so it can

	  be explicitly kept alive as long as the download lasts

	  (regardless of its destruction).

	* src/download_list.cc:

	  Always destroy a download list when destroy() is called, even

	  though "cancel" is a side-effect.

	* src/ui.cc:

	  Don't let mini-buffer downloads overload the Q key.

	* NEWS:

	  Update the NEWS file for 0.3.4 (may get additional updates).

	* po/POTFILES.in:

	  Update the list of files.

	* src/ui.cc:

	  Editorial change to the no-cache-modification-during-a-download

	  message.

	* src/solution_item.cc:

	  Reinstate the advance-after-action behavior of solution items.

	* doc/en/images/Makefile.am:

	  Update the list of distributed images.

	* doc/en/aptitude.xml:

	  More fully document the resolver (the prose is a little stilted

	  in some places but it'll do for a first draft).

	* src/generic/problemresolver/model.tex:

	  Add a disclaimer regarding the stupidity-removal algorithm.

	* src/apt_options.cc:

	  Add an option item for suppress-read-only-warning.

	* src/ui.cc:

	  Make prompt_yesno_popup not always default to 'yes'.

	  (I confused the scrollbar parameter with the default parameter

	  -- arguably this is a sign that vs_dialog_yesno's interface is

	  way too hariy).

	* src/vscreen/vs_multiplex.cc:

	  Bind weak references instead of strong references.

	* src/cmdline/cmdline_update.cc:

	  Fix a silly compile problem.

	* doc/en/aptitude.xml:

	  Fix the documentation as regards the default resolver scores.

	* doc/en/aptitude.xml:

	  Document several undocumented configuration options.

	* doc/en/aptitude.xml:

	  Typo fix.

	* src/cmdline/cmdline_update.cc, src/generic/apt/download_update_manager.cc, src/ui.cc:

	  Add some missing connections to the download managers' hooks.

	* src/cmdline/cmdline_changelog.cc:

	  Kill off some memory leaks.

	* src/changelog_parse.cc:

	  Instead of using the description parser to generate formatted

	  changelogs, just parse them by hand; the bulletting conventions

	  are not, unfortunately, as well-defined as for descriptions.

	* src/menu_text_layout.cc, src/menu_text_layout.h:

	  Make the search keystrokes active in the menu_text_layout

	  widget.

	* src/view_changelog.cc:

	  Use a menu_text_layout instead of a vs_text_layout to display

	  the changelog, and hook it up to the menu signals.

	* src/menu_text_layout.cc, src/menu_text_layout.h:

	  Write a vs_text_layout variant customized to hook the menu's

	  search command up to itself.

	* src/vscreen/vs_text_layout.cc, src/vscreen/vs_text_layout.h:

	  Write basic code for searching in a vs_text_layout.

2005-09-29  Daniel Burrows  <dburrows@debian.org>

	* doc/en/aptitude.xml, src/Makefile.am, src/changelog_parse.cc, src/changelog_parse.h, src/defaults.cc, src/generic/util/temp.h, src/view_changelog.cc:

	  Add support for parsing changelogs and doing something useful

	  with the information therein.  In particular, changelogs are now

	  wrapped like descriptions (and get bullet parsing for free), and

	  newer versions of the package are highlighted.  (Closes: #290692)

	* src/generic/util/temp.h:

	  Allow clients to extract the parent directory of temporary

	  objects.

	* src/cmdline/cmdline_do_action.cc:

	  Fix a nasty bug: when constructing the download manager in

	  do_action, I was actually (and accidentally) constructing a

	  manager on the heap, implicitly casting it to bool (hence always

	  getting 'true') and then passing that as download_only to the

	  true manager.  Obviously this is not quite what was desired.

	* src/ui.cc, src/ui_download_manager.cc, src/ui_download_manager.h, src/view_changelog.cc:

	  Require code that creates a ui_download_manager to set all the

	  usual parameters for its internal download status widget, and

	  change all the creation points to do so (resurrecting the

	  various title strings that got lost earlier).

	* src/cmdline/cmdline_changelog.cc:

	  Fix the command-line changelog code: binding a reference doesn't

	  really work, you have to bind a pointer to the object you want

	  to set instead.

	* src/cmdline/cmdline_changelog.cc, src/view_changelog.cc:

	  Update the changelog viewers for the new interface.

	* src/generic/apt/pkg_changelog.cc, src/generic/apt/pkg_changelog.h:

	  Base the package changelog downloader on the download_manager

	  class.

	* src/Makefile.am, src/download.cc, src/download.h:

	  Remove the old download code.

	* src/cmdline/cmdline_upgrade.cc:

	  Use the download_manager code for upgrades too.

	* src/pkg_item.cc, src/pkg_tree.cc:

	  Eliminate more unnecessary #includes.

	* src/cmdline/cmdline_do_action.cc, src/cmdline/cmdline_update.cc:

	  Eliminate some unnecessary #includes.

	* src/cmdline/cmdline_update.cc:

	  Use the download_manager code in 'update'.

	* src/cmdline_do_action.cc:

	  Use the download_manager code to handle stuff that gets kicked

	  to do_action.

	* src/cmdline_util.cc, src/cmdline_util.h:

	  Write generic code to execute a download object from the

	  command-line.

	* src/ui_download_manager.cc:

	  Actually use both ui_start_download and ui_stop_download.

	* src/ui.cc, src/generic/download_install_manager.cc, src/generic/download_update_manager.cc:

	  Implement the before/after hooks in the UI to shut down curses

	  and to display a message while automatic autocleaning is

	  happening.

	* src/ui.cc:

	  Implement ui_stop_download.

	* po/POTFILES.in, src/Makefile.am, src/ui.cc, src/ui.h, src/ui_download_manager.cc, src/ui_download_manager.h:

	  Split the wrapper class into another file.  Added some public UI

	  routines to allow other code to manipulate the "is something

	  downloading now?" state.

	* src/ui.cc:

	  Instead of subclassing the download managers, use a generic

	  wrapper class that takes a manager as a constructor parameter.

	* src/generic/apt/download_manager.h, src/generic/apt/download_install_manager.h, src/generic/apt/download_install_manager.cc, src/ui.cc:

	  Make prepare() part of the generic download_manager interface

	  (additional arguments should be 'passed' in the constructor

	  now).  This will pave the way for fully generic code in the UI

	  to handle a download.

	* src/generic/apt/download_install_manager.cc, src/generic/apt/download_install_manager.h, src/ui.cc:

	  Make download_only a constructor parameter of the

	  download_install_manager, not an argument of finish().

	* src/generic/apt/download_update_manager.cc:

	  Instead of creating an Acquire object with no status callback,

	  pass our status callback into its constructor; also save the

	  signal generator in signallog rather than just ignoring it.

	* src/ui.cc:

	  First cut at using the download_manager code instead of the old

	  download code.

	* src/download_thread.cc, src/download_thread.h:

	  Accept download managers instead of acquire objects as the

	  thread parameter.

	* src/generic/apt/download_manager.h:

	  Make download managers SigC++ trackable objects.

	* src/generic/apt/download_install_manager.cc, src/generic/apt/download_install_manager.h, src/generic/apt/log.cc, src/generic/apt/log.h:

	  Split the log code from the download code.

	* po/POTFILES.in, src/generic/apt/download_install_manager.cc, src/generic/apt/download_install_manager.h, src/generic/apt/download_manager.cc, src/generic/apt/download_manager.h, src/generic/apt/download_update_manager.cc, src/generic/apt/download_update_manager.h, src/generic/apt/Makefile.am:

	  Write a generic interface to manage operations that involve

	  downloading something and then doing more stuff, and transform

	  the old download code into subclasses of this interface.

	* src/cmdline/cmdline_changelog.cc, src/cmdline/cmdline_do_action.cc, src/cmdline/cmdline_download.cc, src/cmdline/cmdline_progress.cc, src/cmdline/cmdline_progress.h, src/cmdline/cmdline_prompt.cc, src/cmdline/cmdline_update.cc, src/cmdline/cmdline_upgrade.cc, src/download.cc, src/download.h, src/download_list.cc, src/download_list.h, src/download_thread.cc, src/download_thread.h, src/generic/apt/acqprogress.cc, src/generic/apt/acqprogress.h, src/generic/apt/Makefile.am, src/Makefile.am, src/ui.cc, src/ui.h, src/view_changelog.cc:

	  Update include file locations and class names to refer to

	  download_signal_log instead of download_manager.

	* src/download_manager.cc, src/download_manager.h, src/generic/apt/download_signal_log.cc, src/generic/apt/download_signal_log.h:

	  Move the download_manager code into the generic layer and rename

	  it.

2005-09-27  Daniel Burrows  <dburrows@debian.org>

	* src/view_changelog.cc:

	  Since the changelog pager reads in the whole changelog when it's

	  constructed, it no longer needs to ensure that the changelog

	  file (object) exists as long as the pager does; delete the file

	  object as soon as we finish reading it.

	* src/generic/util/exception.h:

	  Update the exception documentation to match reality.

	* configure.ac:

	  Fix the enabling of dynamic backtraces.

	* configure.ac, src/generic/util/exception.cc:

	  Make the dynamic generation of backtraces a configure-time

	  option; in order for them to be useful, a full symbol table

	  needs to be shipped with the program, and in the case of

	  aptitude, this DOUBLES the executable size.

	* configure.ac:

	  Put libraries detected by configure into LIBS, not LDFLAGS.

	* configure.ac, src/generic/util/Makefile.am, src/generic/util/exception.cc, src/generic/util/exception.h, src/main.cc, tests/Makefile.am:

	  Add support for backtracing the program when creating an

	  exception on systems that support it.  (currently rather broken)

	* src/download.cc:

	  Remove the assumption that stdin is line-buffered; explicitly

	  search for a \n character (and handle EOF conditions).

	* src/main.cc, src/cmdline/cmdline_prompt.cc, src/cmdline/cmdline_prompt.h:

	  Toss an exception from the command-line code when EOF is

	  encountered, and write a corresponding catch handler; also

	  replace the special-cased catch blocks in main.cc with generic

	  catches for the Exception type (so the exception gets printed to

	  stderr).  In particular, this means that aptitude no longer goes

	  nuts when it encounters an EOF.  (Closes: #318749)

	* src/ui.cc:

	  Check for the existence of the old temporary directory on

	  startup and display a one-time prompt asking whether it should

	  be removed (currently defaults to "no").

	* src/ui.cc:

	  When su-ing to root, use temp::* instead of making the needed

	  temporary names manually.  (Closes: #146485)

	* src/generic/apt/pkg_changelog.cc, src/generic/apt/pkg_changeloh.h:

	  Use the temp::* classes to manage the changelog instead of

	  manually constructing a temporary name.

	* src/generic/util/Makefile.am, src/generic/util/temp.cc, src/generic/util/temp.h, tests/Makefile.am, tests/test_temp.cc:

	  Add C++-based support code for securely creating and cleaning up

	  temporary files and directories.

	* src/vscreen/config/colors.cc:

	  Don't blow up if no colors are available (e.g., if the terminal

	  doesn't support colors at all).

2005-09-26  Daniel Burrows  <dburrows@debian.org>

	* src/pkg_item.cc:

	  Make the interactive 'keep' command cancel any hold set on a

	  package; the behavior at the command line is unchanged.

	  (Closes: #326949)

	* src/generic/apt/aptcache.cc:

	  Preserve the last version of pkgstates when saving the selection

	  list. (Closes: #316460)  Also, don't remove the .new version if

	  we fail to save the cache (so the user can manually recover).

	* src/ui.cc:

	  Write a better description for the flat package list viewer.

	* src/ui.cc:

	  Also gray out autocleaning and mark-upgradable when the apt

	  cache file isn't available, and don't crash even if someone does

	  call do_autoclean with apt_cache_file == NULL.

	* src/ui.cc:

	  Gray out the install-run menu item whenever the apt cache file

	  isn't available.

	* src/ui.cc:

	  Also clear the 'active download' flag after a download that

	  failed up-front.

	* src/ui.cc:

	  If the download preparation routine fails, tell the download

	  manager that the download was complete so that the download

	  widget is destroyed.

	* src/download.cc:

	  If the user tries to start a download while the package cache is

	  NULL, print a cryptic error message instead of crashing.  If the

	  user tries to start an update while the package cache is NULL,

	  do the update instead of crashing.  (Closes: #309445)

	* src/cmdline/cmdline_moo.cc:

	  Delete the text at the end, which not only lacks a trailing

	  newline, but is also rather rude.  (Closes: #295924)

	* src/broken_indicator.cc:

	  Don't segfault if the apt cache is NULL (e.g., if apt_init

	  failed).  (Closes: #290408) -- at least for the time being; the

	  general problem is that I need to remember to test the program

	  in situations where the apt cache fails to initialize due to,

	  say, invalid configuration files.

	* src/generic/apt/aptcache.cc:

	  Fix the code that sets packages back to manual mode on a keep or

	  hold: it ran after the package was already kept back, so of

	  course the test for whether the package was to be deleted was

	  wrong! (Closes: #278490)

	* src/cmdline/cmdline_action.cc:

	  Only print a message about not being able to forbid an upgrade

	  because the package is not upgradable for packages that are

	  *NOT* upgradable, not packages that are.  (Closes: #246140)

	* src/cmdline/cmdline_show.cc:

	  Abort with an error message if some of the packages the user

	  tried to show don't exist, and return non-zero if errors were

	  generated while showing packages.  (Closes: #301291)

	* doc/en/aptitude.xml, doc/en/manpage.xml, src/generic/apt/apt.cc, src/main.cc:

	  Change Aptitude::Cmdline::Simulate to Aptitude::Simulate; if it

	  is true, always open the cache in read-only mode regardless of

	  whether the user is root or not.  This really Closes: #243192.

	* src/download.cc:

	  Write the date stamps for aptitude logs in RFC2822 format;

	  in particular, add timezones to them.  (Closes: #318501)

	* src/generic/util/util.cc, src/generic/util/util.h:

	  Add a safe wrapper for strerror_r.

	* src/generic/util/util.cc:

	  Correct get_homedir(): it was using the non-array form of delete

	  on arrays (remember, auto_ptr doesn't work on arrays...).

	* src/generic/util/util.cc, src/generic/util/util.h:

	  Write a safe strftime wrapper.

	* src/edit_pkg_hier.cc, src/generic/apt/apt.cc, src/generic/apt/pkg_changelog.cc, src/ui.cc:

	  Consistently use pkg_homedir() instead of getenv("HOME").

	  (Closes: #272429, #274216, #285334)

	* src/generic/util/util.cc, src/generic/util/util.h:

	  Write a utility routine to look up the current user's home

	  directory from the password database.

	* src/main.cc:

	  Automatically activate quiet mode if stdout is not a terminal.

	  (Closes: #276767)

	* src/pkg_item.cc:

	  Hang onto the lock while running reportbug.  This obviates the

	  need to reload the package cache, and in particular it means

	  that aptitude no longer discards information about package

	  states after reporting a bug; hence (Closes: #304748).

2005-09-25  Daniel Burrows  <dburrows@debian.org>

	* src/ui.cc:

	  Add a configuration option to suppress the warning message, and

	  display a customized dialog box that allows the user to disable

	  the message in the future.  With the other changes, this finally

	  (Closes: #175408).

	* src/ui.cc:

	  Make show_message() generate reverse video by default again.

	* src/download_list.cc, src/download_list.h, src/generic/util/slotarg.h, src/ui.h, src/vscreen/slotarg.h, src/vscreen/vs_menu.h, src/vscreen/vs_menu.cc:

	  Move slotarg to the generic code layer.

	* src/generic/apt/aptcache.cc:

	  Add a missing permission check in internal_mark_install.

	* src/ui.cc:

	  Make the package cache read-only while a download is going on.

	* src/vscreen/config/colors.cc:

	  Actually set default_colors_avail based on the result of

	  use_default_colors().  The codepath where this value is false is

	  untested, because I can't seem to find a terminal that doesn't

	  provide default colors.

	* src/ui.cc:

	  Connect up the new cache functions to display a helpful (?)

	  message when a read-only cache is modified.  This enforces a

	  rule that modifying the cache during a download is Not Allowed,

	  and it resets the cache to not-read-only after displaying a

	  single warning message otherwise.

	* src/generic/aptcache.cc, src/generic/aptcache.h:

	  Add support for making the cache 'read-only'; writing to a

	  read-only cache will generate what you could loosely describe as

	  a 'page fault', to be dealt with by slots connected to a public

	  signal.

	* src/generic/util/bool_accumulate, src/ui/ui.h, src/vscreen/bool_accumulate, src/vscreen/vs_menu.h:

	  Move bool_accumulate to generic code.

	* doc/en/aptitude.xml, src/generic/apt/aptcache.cc:

	  Add an option to purge packages that are unused, with big red

	  warning flags in its documentation appropriate to the level of

	  foot-shooting-ness that it enables.  (Closes: #275150)

	* src/cmdline/cmdline_download.cc, src/cmdline/cmdline_upgrade.cc, src/main.cc:

	  Fix the remaining points in the code that need to be updated for

	  quietness.  The code should now work just fine in quiet mode.

	  (Closes: #217477)

	* src/cmdline/cmdline_update.cc, src/cmdline/cmdline_forget_new.cc, src/cmdline/cmdline_progress.cc, src/cmdline/cmdline_do_action.cc:

	  Fix a bunch of typos and update do_action for quietness.

	* src/cmdline/cmdline_clean.cc:

	  Update the clean and autoclean code for quietness.

	* src/cmdline/cmdline_progress.cc:

	  Make downloads quiet.

	* src/cmdline/cmdline_update.cc:

	  Update for quietness.

	* src/cmdline/cmdline_resolver.cc, src/cmdline/cmdline_spinner.cc, src/cmdline/cmdline_spinner.h:

	  Disable output from the spinner when the quiet level is greater

	  than zero.

	* doc/en/aptitude.xml, doc/en/manpage.xml, src/main.cc:

	  Add basic support for a 'quiet' option; it needs to be

	  integrated into all cmdline files.

	* doc/en/aptitude.xml:

	  Explicitly note that ~a matches the *current state* of a

	  package; for instance, ~aupgrade doesn't do what ~U does.

	  (Closes: #311290)

	* FAQ, doc/en/aptitude.xml:

	  Apply several typo fixes submitted by

	  Olivier Trichet <olivier.trichet@freesurf.fr>.

	  (Closes: #268916)

	* doc/en/manpage.xml:

	  Fix the SEE ALSO reference to the full documentation

	  (Closes: #265723).

	* src/vscreen/vscreen.cc:

	  Don't initialize the unused style WidgetFrame.

	* doc/en/aptitude.xml, src/defaults.cc, src/ui.cc:

	  Add keybindings and menu items to directly select the first and

	  last generated solution.

	* src/load_config.cc, src/vscreen/config/colors.cc, src/vscreen/config/colors.h, src/vscreen/config/style.cc, src/vscreen/config/style.h:

	  Support using the default background color by encoding it in the

	  slots that would be used for an identical foreground &

	  background color.  Attempts to set identical foreground &

	  background now return an arbitrary color with the same

	  background, and the style code sets A_INVIS in such a case

	  (resulting in a blank space with the same background).  This is

	  a bit of a hack, but it's the only way to do this when you

	  consider that xterm provides 9 colors (including the default

	  color) but only 64 color pairs.  (Closes: #161872)

	* src/cmdline/cmdline_clean.cc:

	  Unreverse the logic that determines the return value of 'clean':

	  it was returning 0 iff there WERE errors.  (Closes: #274098)

	* src/ui.cc:

	  Try to start a background solution calculation when connecting

	  up the signal (it's possible that we missed a signal that gets

	  emitted when the resolver is created).

	* src/cmdline/cmdline_upgrade.cc:

	  When we have to abort an upgrade, display any broken packages

	  that forced us to abort. Maybe this should hook into the full

	  resolver in the future? (Closes: #316027)

2005-09-25 Miroslav Kure <kurem@upcase.inf.upol.cz>

	* update Czech translation. Closes:#330014

2005-09-24  Daniel Burrows  <dburrows@debian.org>

	* src/generic/util/setset.h, tests/***:

	  Fix more include files that I missed (oops).

	* src/vscreen/vs_menubar.cc:

	  Display left and right arrows at the edge of the terminal if the

	  menu bar is too large for it.

	* src/***:

	  Fix all the include file problems and other stuff arising from

	  the reorganization.

	* po/POTFILES.in:

	  Update POTFILES.in.

	* Makefile.am, src/Makefile.am, src/vscreen/Makefile.am, src/generic/***:

	  Massively reorganize the generic/ subdirectory.  The code is now

	  split into three parts: code which is fully generic (util/), the

	  problem resolver, and "generic" apt routines (apt/).  Not only

	  is this more aesthetically pleasing, it also means that there

	  are no weird cross-directory dependencies (dependencies still

	  exist but they aren't weird any more: specifically, a straight

	  "make" will now always build code before its dependencies are

	  built.

	* src/ui.cc:

	  Fix do_next_solution_enabled().

	* src/generic/resolver_manager.cc, src/generic/resolver_manager.h:

	  Eliminate out_of_solutions; use the solutions_exhausted member

	  of the state snapshot to calculate

	  solution_generation_complete().

	* src/generic/resolver_manager.cc:

	  Use the resolver counts instead of out_of_solutions to set the

	  solutions_exhausted member of the state snapshot.  This fixes

	  the 'are there more solutions?' indicator.

	* src/generic/problemresolver/problemresolver.h:

	  Add an element to the counts cache that mirrors 'finished', and

	  update the cache at the important exit points from the solver

	  and (if no solver is running) when retrieving queue counts.

	* src/generic/problemresolver/problemresolver.h:

	  When screening out solutions that we've already generated a

	  subset of, take into account the set of unresolved dependencies

	  as well as the set of actions.

	* src/generic/problemresolver/solution.h:

	  If an action coincidentally fixes an unresolved dependency,

	  remove that dependency from the unresolved set.

	* src/generic/problemresolver/problemresolver.h:

	  Remove the forced_broken interface from the resolver itself.

	  Switch a call to unforce_break that I missed earlier to invoke

	  unapprove_break instead.

	* src/generic/resolver_manager.cc, src/generic/resolver_manager.h:

	  Rip the forced_broken stuff out of the resolver manager.

	* src/solution_item.cc:

	  Switch the solution_unresolved_item class to using the new

	  approved_broken stuff instead of forced_broken.

	* src/generic/problemresolver/problemresolver.h:

	  Doh, of course the MUTATORS for the new structure shouldn't be

	  constant!

	* src/generic/resolver_manager.cc, src/generic/resolver_manager.h:

	  Expose the approved_broken stuff from the manager.

	* src/generic/resolver_manager.cc, src/generic/resolver_manager.h:

	  Collect all the resolver manipulations into a common routine.

	* src/generic/problemresolver/problemresolver.h:

	  Add a set of "approved breaks" that should provide a more

	  sensible way of handling soft dependencies: instead of forbidden

	  the dependency to be solved outright, it just says that we

	  should place it in the 'unresolved' set whenever it comes up as

	  an option.  In other words, it defers any solution in which an

	  action is labelled with that dependency.

	* src/generic/problemresolver/problemresolver.h:

	  Fix the resolver indentation (finally).

2005-09-24 Rubén Porras <nahoo@inicia.es>

	* Update spanish translation

	* regenerate po/aptitude.pot, po/POTFILES.in and po files

2005-09-23  Daniel Burrows  <dburrows@debian.org>

	* src/generic/problemresolver/problemresolver.h:

	  When force-breaking a soft dependency, make sure to reject its

	  source as well.  NOTE: the semantics of force-breaking may be a

	  bit off; maybe they should be reviewed to better match approval?

	  (i.e., only reject solutions due to force-breaking if something

	  force-broken is listed as a justification)

	* src/ui.cc:

	  Connect state_changed to vscreen_update, so an update is always

	  enqueued when the resolver state changes.  In particular, this

	  ensures that a keyboard-invoked undo will always trigger a

	  screen update.

	* src/generic/resolver_manager.cc:

	  For safety, always unsuspend a background thread before calling

	  state_changed().

	* src/generic/resolver_manager.cc:

	  Fix background_suspender::unsuspend -- it wasn't clearing its

	  suspended flag, so the suspender would try to unsuspend AGAIN

	  when it was destroyed.

	* src/solution_screen.cc:

	  Write a solution_undo_tree (analogue of apt_undo_tree) and use

	  it to activate the undo command in solution displays.

	* src/ui.cc:

	  Remove the no-longer-necessary do_undo() and undo_enabled()

	  routines.

	* src/apt_info_tree.h, src/pkg_tree.h:

	  Derive apt_info_tree and pkg_tree from the new undo-supporting

	  tree.

	* src/apt_undo_tree.cc, src/apt_undo_tree.h:

	  Add a tree class that implements an apt undo interface (i.e., so

	  package manipulation screens work as expected).

	* src/menu_redirect.cc:

	  Fix a rather silly typo.

	* src/menu_redirect.cc, src/menu_redirect.h:

	  Add menu_redirect hooks for the "undo" command.

	* src/ui.cc, src/ui.h:

	  Connect undo() through a signal instead of directly.  For now

	  do_undo() is still used, though.

	* src/generic/resolver_manager.cc:

	  Always signal state_changed() after an undo.

	* src/generic/resolver_manager.cc, src/generic/resolver_manager.h:

	  Add support for tracking an undo stack for a resolver manager's

	  state.

	* src/generic/problemresolver/problemresolver.h, src/generic/problemresolver/resolver_undo.h:

	  Correct a typo (generic_problem_resolver, not generic_resolver)

	  so the new code compiles.

	* src/generic/problemresolver/problemresolver.h, src/generic/problemresolver/resolver_undo.h:

	  Add support in the resolver for recording the inverse of

	  operations performed on it.  (could be done at a higher level,

	  but for obvious reasons, the resolver is uniquely positioned to

	  know exactly how to reverse its actions)

	* src/vscreen/vs_menu.cc, src/vscreen/vs_menu.h:

	  Allow the user to scroll the menu up and down on small

	  terminals.

	* src/view_changelog.cc:

	  Write a long menu description for the changelog pager.

	* src/ui.cc:

	  Edit the long menu description of the manual pager.

	* src/ui.cc:

	  Display the user's manual as a top-level widget instead of a

	  dialog box.

	* src/solution_screen.cc:

	  Switch back to using the widget itself as the menu target, and

	  bind menu handlers for the info area; the resolver menu works

	  now!

	* src/menu_redirect.cc:

	  Merge do_menu_test and do_menu_action into do_menu_callback, and

	  make the new routine use get_widget_active() to filter calls:

	  menu actions on a widget are valid only when every ancestor of

	  the widget is either the root or the 'active' widget of its

	  owner.

	* src/vscreen/vs_multiplex.cc:

	  Make it compile.

	* src/vscreen/vscreen_widget.cc, src/vscreen/vscreen_widget.h:

	  Make get_owner() return a ref_ptr, not a bare pointer.

	* src/vscreen/vs_container.h, src/vscreen/vs_menubar.cc, src/vscreen/vs_menubar.h, src/vscreen/vs_passthrough.cc, src/vscreen/vs_passthrough.h:

	  Add a routine to fetch the 'active' widget of a container (where

	  the definition of 'active' varies by widget, but is usually the

	  widget with the interaction focus).

	* src/ui.cc:

	  Add shortcut characters to the Resolver menu.

2005-09-22  Daniel Burrows  <dburrows@debian.org>

	* src/menu_tree.cc:

	  Say "Search Backwards", not "Search Back" (more readable).

	* src/ui.cc:

	  Don't talk about packages in the long menu description of

	  Search->Find and Search->Find Backwards now that other things

	  support those commands.

	* src/view_changelog.cc, src/view_changelog.h:

	  Add support for either forward or backwards searching.

	* src/vscreen/vs_pager.cc, src/vscreen/vs_pager.h:

	  Add support for searching backwards in a file.

	* src/menu_redirect.cc, src/menu_redirect.h, src/pkg_node.cc, src/solution_item.cc:

	  Write default implementations for the menu_redirect methods that

	  cause all commands to be disabled; avoids lots of repetitive and

	  pointless code in subclasses.

	* src/view_changelog.cc:

	  Make sure that the changelog pager's scroll bar occupies the

	  full vertical extent of the screen.

	* src/menu_tree.cc, src/menu_tree.h, src/pkg_item.cc, src/pkg_item.h, src/pkg_node.cc, src/pkg_node.h, src/pkg_ver_item.cc, src/pkg_ver_item.h, src/solution_item.cc, src/solution_item.h:

	  Split up the menu_tree monstrosity and eliminate most of the

	  dynamic_casts.  Now it just does a single dynamic_cast of the

	  currently selected item to the menu_redirect interface; if the

	  cast succeeds, it just proxies for the result.

	* src/solution_screen.cc:

	  D'oh, connect up the bindings to the examiner, not the info

	  tree.

	* src/menu_tree.cc, src/menu_tree.h:

	  Implement the resolver commands.  This exposed some real

	  kludginess in the menu_tree that should be fixed in the future

	  (I imagine it acting as simply a proxy for the selection when

	  the selection is a menu_redirect).

	* src/solution_item.cc, src/solution_item.h:

	  Split the key response actions into public methods.

	* src/Makefile.am, src/apt_info_tree.h, src/load_grouppolicy.cc, src/menu_tree.cc, src/menu_tree.h, src/pkg_tree.cc, src/pkg_tree.h, src/solution_screen.cc, src/ui.cc:

	  Rename pkg_menu_tree to menu_tree and split it into a separate

	  file from pkg_tree.

	* src/solution_screen.cc:

	  Use a pkg_menu_tree instead of a vs_tree to build the solution

	  screen.  Note that pkg_menu_tree is a bit of a misnomer now...

	* src/pkg_tree.h:

	  Make pkg_menu_tree::create a static method (whoops).

	* src/menu_redirect.cc:

	  Bind up the resolver menu signals.

	* src/Makefile.am, src/menu_redirect.cc, src/menu_redirect.h, src/pkg_view.cc:

	  Split the code to bind up a menu_redirect object and the active

	  package view out into a separate .cc file.

	* src/ui.cc:

	  Actually put the new items in the menu and define the signals.

	* src/ui.h:

	  Add signals that will be used to hook up the new menu items.

	* src/solution_item.cc, src/solution_item.h, src/solution_screen.cc:

	  Split the tree items from solution_screen.cc into their own

	  file.

	* src/ui.cc:

	  Typo fix.

	* src/menu_redirect.h, src/pkg_tree.cc, src/pkg_tree.h:

	  Add stubs for future resolver menu items.

	* src/ui.cc:

	  Add a Resolver menu with various resolver commands, for those

	  who like such things.

	* src/vscreen/vs_staticitem.cc:

	  Make vs_staticitem really use the whole width (it was using one

	  character less).

	* src/solution_screen.cc:

	  Add more verbiage to unresolved soft deps when they show up as

	  possible alternatives to the selected action.

	* doc/en/aptitude.xml, src/defaults.cc, src/solution_screen.cc:

	  Modify the SolutionActionAccepted style and the

	  SolutionActionAccept keybinding to conform to the new "approved"

	  terminology; don't overspecify the SolutionAction styles.

	* doc/en/aptitude.xml, doc/en/images/broken-snapshot.png, doc/en/images/solution-screen-snapshot.png, doc/en/images/solution-screen-snapshot2.png, doc/en/images/solution-screen-snapshot3.png, doc/en/images/solution-screen-snapshot4.png, doc/en/images/solution-screen-snapshot5.png:

	  Update the broken packages tutorial.

	* src/generic/problemresolver/problemresolver.h, src/generic/resolver_manager.cc:

	  Make approving and rejecting an action mutually exclusive:

	  actions that are approved are automatically un-rejected, and

	  vice versa.

	* src/generic/problemresolver/problemresolver.h:

	  Allow mandated versions to be ignored in favor of other mandated

	  versions, and don't special-case other versions of the package.

	* doc/en/aptitude.xml, src/defaults.cc:

	  Don't explicitly set the background color in package colors, so

	  the user can globally override the background color if he/she so

	  desires.

	* doc/en/aptitude.xml, src/defaults.cc, src/pkg_item.cc, src/pkg_item.h, src/reason_fragment.cc:

	  Make a separate named style for each and every possible package

	  item state, including both highlighted and non-highlighted

	  styles.  Among other things, this makes xterm display reasonable

	  cursor colors for highlighted packages (by removing the

	  A_REVERSE flag, which seems to confuse it in some cases).  Of

	  course, it also means that people have much more flexibility in

	  customizing the colors of the package list.

	* src/generic/resolver_manager.cc:

	  Optimize for the case of block_steps == 0 (so users can

	  stop it from using any threading primitives at all).

	* src/vscreen/vs_menubar.cc, src/vscreen/vs_menubar.h:

	  If the screen is too narrow to display a whole menubar, allow it

	  to "scroll" while the user is interacting with it.

	* src/vscreen/vscreen.cc:

	  (hopefully) fix the cursor placement problems that I was having.

	* src/vscreen/curses++.h:

	  Fix mvaddnstr for std::wstrings.

2005-09-22 Rubén Porras <nahoo@inicia.es>

	* Update spanish translation

	* regenerate po/aptitude.pot and po files

2005-09-21  Daniel Burrows  <dburrows@debian.org>

	* doc/en/aptitude.xml:

	  Document Aptitude::ProblemResolver::WaitSteps and

	  Aptitude::Spin-Interval.

	* src/generic/resolver_manager.cc:

	  Don't hold the lock while calling state_changed().  This avoids

	  deadlocks if state-changing callbacks trigger code that tries to

	  grab the instance lock from another thread.  (and just avoids

	  holding the lock over arbitrary code in general, probably a good

	  idea)

	* src/ui.cc:

	  Use the new maybe-blocking solution fetching routine.

	* src/generic/resolver_manager.cc:

	  Add support for cancelling a suspension before destroying the

	  corresponding background_suspender object.

	* src/generic/resolver_manager.cc, src/generic/resolver_manager.h:

	  Add a routine that waits for the resolver to perform some number

	  of steps before returning; can be used to avoid displaying a

	  progress indicator unless the search seems to be taking a while.

	* src/ui.cc:

	  Ignore attempts to apply solutions that don't exist yet.

	* src/broken_indicator.cc, src/cmdline/cmdline_resolver.cc, src/generic/resolver_manager.cc, src/generic/resolver_manager.h, src/solution_dialog.cc, src/solution_screen.cc, src/ui.cc:

	  Make the number of steps an explicit parameter to get_solution

	  and get_solution_background.

	* src/ui.cc:

	  Signal that the resolver state changed when it runs out of

	  solutions.

	* src/broken_indicator.cc:

	  Fix the placement of a bunch of code in broken_indicator::update

	  that accidentally ended up inside a loop.

	* src/broken_indicator.cc:

	  Dim out the unavailable actions in the broken hint line.

	* src/vscreen/fragment.cc:

	  Make column fragments always have a final newline.

	* src/vscreen/vscreen.cc:

	  When determining how long the timeout thread should sleep,

	  return the absolute time that the next timeout will trigger at,

	  not the amount of time until that time (this results in the

	  timeout thread spinning forever).

	* src/ui.cc, src/ui.h:

	  Trigger package_states_changed from apply_current_solution.

	  Among other things, this means that the information area is

	  properly updated when a solution is applied.

	* src/solution_screen.cc:

	  Destroy the whole examiner when there aren't any broken

	  packages, and do it from external code (not update() or a

	  connection to destroyed) to avoid any possibility of weirdness.

	* src/cmdline/cmdline_do_action.cc, src/generic/aptcache.cc, src/generic/aptcache.h, src/ui.cc:

	  Upgrade packages that were to-be-removed when doing a

	  dist-upgrade from the command line.  This is achieved by adding

	  a new flag to mark_all_upgradable that selects one or the other

	  behavior; also fixed the handling of Unknown packages in

	  mark_all_upgradable (it was just wrong before, and probably

	  wasn't noticed because that case should never come up anyway).

	* src/broken_indicator.cc, src/dep_item.h, src/download_bar.h, src/download_list.h, src/edit_pkg_hier.h, src/mine/cmine.h, src/pkg_info_screen.h, src/pkg_tree.h, src/pkg_ver_item.h, src/pkg_view.cc, src/solution_dialog.cc, src/solution_screen.cc, src/ui.cc, src/view_changelog.cc, src/vscreen/vs_button.h, src/vscreen/vs_center.h, src/vscreen/vscreen_widget.h, src/vscreen/vs_editline.h, src/vscreen/vs_frame.h, src/vscreen/vs_label.cc, src/vscreen/vs_label.h, src/vscreen/vs_menubar.h, src/vscreen/vs_menu.h, src/vscreen/vs_minibuf_win.h, src/vscreen/vs_multiplex.h, src/vscreen/vs_pager.h, src/vscreen/vs_scrollbar.h, src/vscreen/vs_size_box.h, src/vscreen/vs_stacked.h, src/vscreen/vs_statuschoice.h, src/vscreen/vs_table.h, src/vscreen/vs_text_layout.h, src/vscreen/vs_togglebutton.h, src/vscreen/vs_transient.h, src/vscreen/vs_tree.h:

	  Make the reference count of vscreen_widgets default to 1, not 0.

	  As a result, it is necessary for every creation routine to

	  explicitly decref() its return value before returning it.  Also

	  wrote a ::create() routine for cmine (it was missing one) and

	  changed ui.cc to call it.

	  The reason for doing this is that it allows routines run from

	  the constructor to acquire and release references without

	  causing the world to explode violently.

	* src/apt_info_tree.cc, src/broken_indicator.cc, src/download_bar.cc, src/download_list.cc, src/download_screen.cc, src/edit_pkg_hier.cc, src/mine/cmine.cc, src/pkg_info_screen.cc, src/pkg_view.cc, src/solution_dialog.cc, src/solution_screen.cc, src/vscreen/vs_bin.cc, src/vscreen/vs_button.cc, src/vscreen/vs_center.cc, src/vscreen/vs_container.cc, src/vscreen/vscreen_widget.cc, src/vscreen/vs_editline.cc, src/vscreen/vs_frame.cc, src/vscreen/vs_label.cc, src/vscreen/vs_menubar.cc, src/vscreen/vs_menu.cc, src/vscreen/vs_minibuf_win.cc, src/vscreen/vs_multiplex.cc, src/vscreen/vs_pager.cc, src/vscreen/vs_passthrough.cc, src/vscreen/vs_scrollbar.cc, src/vscreen/vs_size_box.cc, src/vscreen/vs_stacked.cc, src/vscreen/vs_statuschoice.cc, src/vscreen/vs_table.cc:

	  Take a self-reference in any routine that's not totally trivial

	  (in particular, any routine that invokes a virtual method or a

	  signal) in order to avoid future surprises like the one that

	  wasted so much time yesterday.

2005-09-20  Daniel Burrows  <dburrows@debian.org>

	* src/apt_info_tree.cc:

	  Preemptively fix a likely problem that turned up in an audit for

	  other instances of the access-this-after-self-destructing bug.

	* src/solution_screen.cc:

	  Have the examiner take a reference to itself in the destroy()

	  call; fixes a crash due to accessing members of the examiner

	  after it's deleted by destroy().

	* src/solution_screen.cc:

	  Clear out the other information areas when the multiplex goes.

	* src/vscreen/vs_multiplex.cc, src/vscreen/vs_table.cc:

	  Tighten some asserts.

	* src/vscreen/vs_table.cc:

	  Fix the logic that chooses a new focus after the current focus

	  goes away: most places were only looking at *UN*focussable

	  widgets; instead, only consider widgets that are visible and can

	  be focussed.

	* src/vscreen/vscreen_widget.cc:

	  Don't mark the widget as having been destroyed until after it's

	  been hidden and removed from its owner; this avoids some nasty

	  weirdness due to the way that the vscreen_widget accessors

	  change their behavior when the widget is destroyed.

	* src/broken_indicator.cc, src/solution_screen.cc:

	  Display the resolver progress in the solution screen instead of

	  in the broken indicator; use the new slot timeout object

	  to set up the broken indicator's timeout.

	* src/vscreen/vscreen.cc, src/vscreen/vscreen.h: