January 2007 - Pacman-dev - lists.archlinux.org

[pacman-dev] [patch] Preliminary cleaning of NEWS file
by Dan McGee 12 Jan '07

12 Jan '07

Quick look through getting rid of bugfixes that occurred in never-released versions on the Arch side. Added a few TODO's. Signed-off-by: Dan McGee <dpmcgee(a)gmail.com> --- --- pacman-lib.orig/NEWS 2007-01-07 21:11:03.000000000 -0500 +++ pacman-lib/NEWS 2006-12-21 13:55:56.000000000 -0500 @@ -1,14 +1,13 @@ VERSION DESCRIPTION ----------------------------------------------------------------------------- +FUTURE - drop the NoUpgrade lines from the default config 3.0.0 - TODO this needs alot of work + *group translation/i18n stuff together + *remove anything that isn't applicable between unreleased versions - switched to libfetch in place of libftp - attempt to speed up file reading by reading /desc and /depends on demand - - bugfix for holdpkgs during an upgrade - - fix building with --enable-debug - fixed the download progressbar - - -Ss works again for multiple repos - - bugfix for -Sp - downloads are interruptable again - new libalpm function: alpm_db_search() - pactest: initial import, run make check to call it @@ -23,185 +22,67 @@ - makepkg: recoginze .TGZ, .ZIP and other uppercase extensions - python: added a few casting functions - ported libalpm to Solaris/Sparc - - bugfixes in the Hungarian translation - - bugfix for pacman -Qi -p - new function in libalpm: alpm_db_setserver() - incremented libalpm's sover due to new error codes intruduced by alpm_db_setserver() - added python bindings - - fixed a stale lockfile problem (when packages were corrupted) - fixed problem when sometimes ^C was ignored - new pacman option: --ask - libalpm: prefixed all internal functions with _alpm - makepkg: bugfix for suboptions=('nostrip') - - new package property: "build type" - it can be "chroot" or - custom - updated Hungarian translation - - the old extra repo is now merged to the frugalware one - handle the case properly when the user hits ^C during a package install - - fix the output of scriptlet messages - check if enough free space if available - - pipe and parse scriptlet messages. see man FrugalBuild - - new mirror: ftp10 - - udatesync: use bzip2 instead of gzip for package databases - - update/gensync: bugfix for the pkgdesc_localized directive - - Hungarian translation update - - fixed several compilation warning - ia64 support - - manpage split: moved the FrugalBuild specific part from - man makepkg to man FrugalBuild - - minor bugfixes in makepkg - gettext improvements - French translation - - makepkg support and documentation for pkgdesc localization - pacman: display uncompressed package size by default on upgrade - initial support for pkgdesc localization - - bugfix for -Qi and -Qm - - fix -Sg for multiple repos - - drop the NoUpgrade lines from the default config - - makepkg: when removing the pkg dir, remove the subpkgs' pkg - dir, too - - significant speedup while sorting dependencies with libalpm - fixes for pacman's -r option in pacman and libalpm - initial support for uncompressed package sizes in a sync repo - - minor fixes for de_DE, hu_HU and tr_TR - - seems that the original pacman3's release is delayed so it's - no good in waiting for it. changed the versioning scheme to - our usual x.y.z - - i18n fixes: - in the german translation - in makepkg - - gpg support in makepkg -2.8.99.28 - added gettext support + - added gettext support - added German and Hungarian translation for libalpm and pacman -2.9.99.27 - scripts: cleanup, they now require util.sh from FST - makepkg: handle spaces in source filenames - - makepkg: spelling fixes - - libalpm: fixed a few messages' verbosity level - - ./autogen.sh --dist now allows to build a snapshot fpm -2.9.99.26 - makepkg: new option -t (to be used with -R) - added i386, OpenBSD and OS X support - - updated mirror list - - spelling fixes - - bugfix in pacman (dbpath initialization) -2.9.99.25 - bugfix in the conflicts/provides handling code - - spelling fix in makepkg - - include the root string (ie. a slash) in file conflict error - messages - - gensync: bugfix for subpkgs -2.9.99.24 - fixed the location of the locally installed packages' - install/changelog file (so that pre/post_remove and -Qc works - perfectly again) - allow repo names to be longer than 128 chars - - fixed pacman -v output (not a bug just it was ugly) -2.9.99.23 - bugfix for -Sy - - gensync: ignore _darcs/ directories -2.9.99.22 - new feature: options=('scriptlet') to force executing the - install scriptlets even if one builds in a chroot - - makepkg -RC now uses the chroot's pacman.conf to determine - what packages need to be removed. this is especially useful - if you use unofficial repos only in chroot - the testsuite is replaced by pactest, see HACKING -2.9.99.21 - bugfix release -2.9.99.20 - handle --noprogress on install/upgrade, too + - handle --noprogress on install/upgrade, too - disable progressbar by default if the output is redirected - - offer deleting a packages if it is corrupted - - makepkg: install subpackages with -i, too - in chroot sync /etc/pacman.d/* from the host system - use the chroot's pacman.conf when it's available - - new pacman option: --noscriptlet for disabling install - scriptlets - - new variable in makepkg: $FCC in case the package would use - gcc directly (not from $PATH) -2.9.99.19 - more verbose error messages on corrupted packages - - makepkg: added support for clearing the cache of ccache - slice warnings for LANG!=C - - gen/updatesync: bugfix for subdescs() and subarchs() - no longer ignore the Finclude function call -2.9.99.18 - makepkg now allows to inherit install scriptlets from a - scheme + - more verbose error messages on corrupted packages - allow multiple packages to be installed at once with -U - - added SBU variable to makepkg.conf - - 3 major bugfixes from upstream -2.9.99.17 - support for having subpkgs in an other repo. this affects - gensync and updatesync, also makepkg's manpage is updated - support for handle version comparators in makepkg - - bugfix for Finclude in makepkg - removed exit() calls from libalpm - - bugfix for nobuild packages in sync -2.9.99.16 - fixed a possible infinite loop during -S - - remove stale lockfile when pacman is interrupted - - new directive: added suboptions() and subarchs() for - FrugalBuilds - - added 'nofakeroot' to options() - - new makepkg witch: -RCC, now -RC only cleans the chroot -2.9.99.15 - fix 'error: transaction already initialized' when using -D -2.9.99.14 - documentation: improved libalpm manpages an added a main + - documentation: improved libalpm manpages an added a main libalpm(3) manpage - - yet another bugfix for the install/upgrade progressbar - - remove stale lockfile after -R/-A - - fixed a segfault when removing a package that is reqired by - other packages -2.9.99.13 - Allow queries (-Q, -Si, etc) during an other pacman is - running (ie. -Syu) - - Two bugfixes in connection with provides - - Improved makepkg output for subpackages (crazy) -2.9.99.12 - fixed preparing docs, 2.9.99.11 always supposed we want to - disable that feature -2.9.99.11 - fixed progressbar when installing and upgrading packages at + - fixed progressbar when installing and upgrading packages at the same time - - in general, modified makeworld to work properly on Frugalware - - makepkg: if build fails, save the buildlog when using -R - sudo improvements for makepkg & makeworld - display "checking for conflicting files..." during -S - initial doxygen support -2.9.99.10 - fixed -Sg output - now frugalpkg works again - - fixed a double free() when installing conflicting packages - - fixed an annoying bug with progressbar, now works as it did - in the pacman 2.x times - don't care about a to-be-replaced package's newer version -2.9.99.9 - fixed -S/-U/-A segfault. hopefully not everybody upgraded - yet... -2.9.99.8 - if you have foo installed, and bar replaces + provides foo, + - if you have foo installed, and bar replaces + provides foo, then remove foo, and install bar (now works again) - upgrade only pacman first during a -Su - - removed duplicated error messages - more verbose error message for conflicting files when using -S -2.9.99.7 - bugfix for -Sg - if pacman can't create its cachedir, then use /tmp (bugfix) -2.9.99.6 - bugfix for -Rs - - pre_install/pre_upgrade scripts are working again - - added a -f option to gensync/updatesync (don't die if the fpm - is missing) - - bugfix for --ignore - added ppc support (to configure) - handle broken databases (just print an error message, don't segfault) - more verbose error messages for unsatisfied dependencies -2.9.99.5 - re-added SBU support to makepkg (which was accidently dropped - by 2.9.99.0) - - initial package splitting support - - memleak fixes - more verbose error messages on file conflicts -2.9.99.4 - fixed running scriptlets when -r is used - - don't die on up to date databases (when you do -Sy twice) - more verbose error messages (ie. when there is a missing dependency) -2.9.99.3 - fixed -v option + - fixed -v option - allow c++ usage in alpm.h - - fixed logging everything twice to pacman.log -2.9.99.2 - makepkg -R works again if you haven't had a chroot yet (was a - pacman bug) + - fixed logging everything twice to pacman.log TODO: dan- i think i saw this happen when i installed things - output fix for -Sy - if you have installed foo, and then you do a -S bar, then now pacman offers you to remove foo (as it should) -2.9.99.1 - fix makepkg -g - makepkg: fixed changelog generation - fixed --noconfirm -2.9.99.0 - this is the first release which is based on pacman-lib - (converted the initial cvs repo to darcs) + - change to pacman-lib based version 2.9.8 - Changed behaviour with original=X,current=Y,new=Z scenario - keep old in place, install new as .pacnew - Search package provides when finding matching targets with -S

4 5

[pacman-dev] pacman3 logging issues
by Dan McGee 11 Jan '07

11 Jan '07

Just thought I'd send this email before I forgot about it. Here is a log from my pacman3 -Syu yesterday: http://pastebin.archlinux.org/548 The aspell upgrade alone was logged 9 times...so I'm thinking there is a bit of an issue here with the way log entries are being written to the file. I haven't looked at the logging code enough yet to know what is going on, but I thought I'd at least present a concrete example to the list. Happy holidays to everyone! -Dan

3 5

[pacman-dev] New test install - bug fixes
by Aaron Griffin 11 Jan '07

11 Jan '07

Closed out a few bugs and threw up a new version. This includes all the recent script changes (and is built with the new makepkg). Important bug fixes: * --print-uris works again (thus, so does checkpkg, for the devs) * Roman's dep corruption (major bug, minor fix, heh) * Few others.... http://archlinux.org/~aaron/pacman/pacman-rc-3.0.0-3-i686.pkg.tar.gz If you can, please test the bugs listed here that say "Requires Testing": http://bugs.archlinux.org/index.php?project=1&type=1&cat=6 Thanks guys, Aaron

2 1

[pacman-dev] makeworld script- anyone use it?
by Dan McGee 11 Jan '07

11 Jan '07

I've noticed the makeworld script is in the pacman CVS, and I've never used the program in my life. Does anyone still use this, and is it worth updating? This, to me, seems to almost violate the KISS principle and should be spun off into another package (like srcpac is). Any thoughts appreciated. -Dan

6 6

[pacman-dev] [PATCH] _alpm_db_load_pkgcache: use mergesort to improve performance
by Jürgen Hötzel 11 Jan '07

11 Jan '07

Hi, Yesterday i started to analyze pacman's internal. pacman's db speed is typically bound by your systems IO performance: It has to read all db files, if you search repositories. But i was also surprised by the CPU time utilized. So i started profiling "pacman -Ss test": calls ms/call ms/call name 3751 0.01 0.01 _alpm_list_add_sorted 2088191 0.00 0.00 _alpm_pkg_cmp _alpm_list_add_sorted is called 2088191 times while building the pkgcache. This is because every package is sorted into a linked list while scanning the database files: while((info = _alpm_db_scan(db, NULL, infolevel)) != NULL) { ... db->pkgcache = _alpm_list_add_sorted(db->pkgcache, info, _alpm_pkg_cmp); ... Because on each insert the list has to be traversed, pkgcache building has time complexity of O(n^2). Merge-sort is the quickest sort algorithm for linked lists (O(n*log(n)). So i changed above code to use merge-sort. Results on my System: bash-3.2$ time ./src/pacman/pacman.static.old -Ss test >/dev/null real 0m0.791s user 0m0.432s sys 0m0.228s bash-3.2$ time ./src/pacman/pacman.static -Ss test >/dev/null real 0m0.493s user 0m0.232s sys 0m0.260s bash-3.2$ Well, only 0.2 seconds for 70 lines of code patch ;) The different complexity becomes obvious, if you use large repositories. I created a 15000-dummy-packages repository: bash-3.2$ time ./src/pacman/pacman.static.old -Ss test >/dev/null real 0m5.530s user 0m4.588s sys 0m0.828s bash-3.2$ time ./src/pacman/pacman.static -Ss test >/dev/null real 0m2.127s user 0m0.836s sys 0m1.032s Don't be surprised by the small real time. I started each command multiple to prevent IO. Jürgen

2 1

[pacman-dev] [patch] remove configure check for bzip2
by Dan McGee 11 Jan '07

11 Jan '07

Switch to libarchive means this check should no longer be needed. Signed-off-by: Dan McGee <dpmcgee(a)gmail.com> --- --- pacman-lib.orig/configure.ac 2007-01-07 21:11:03.000000000 -0500 +++ pacman-lib/configure.ac 2007-01-08 10:55:01.000000000 -0500 @@ -291,14 +291,6 @@ JAVASTATUS="no, disabled by configure" fi -dnl Check for bzip2 -AC_CHECK_LIB([bz2], [BZ2_bzCompress], [AC_CHECK_HEADER([bzlib.h], [LIBBZ2='-lbz2'])]) -if test -n "$LIBBZ2"; then - LDFLAGS="$LDFLAGS $LIBBZ2" -else - AC_MSG_ERROR("bzip2 library and its include files are needed to build pacman!"); -fi - dnl Check for math AC_CHECK_LIB([m], [log10], [AC_CHECK_HEADER([math.h], [LIBM='-lm'])]) if test -n "$LIBM"; then

2 1

[pacman-dev] [patch] alpm documentation updates
by Dan McGee 11 Jan '07

11 Jan '07

There is still a lot more to do...someone on this list wants to step up to the plate and fix up the documentation. Signed-off-by: Dan McGee <dpmcgee(a)gmail.com> --- --- pacman-lib.orig/lib/libalpm/alpm.c 2007-01-03 03:05:13.000000000 -0500 +++ pacman-lib/lib/libalpm/alpm.c 2007-01-07 23:23:26.000000000 -0500 @@ -66,7 +66,7 @@ pmhandle_t *handle = NULL; enum __pmerrno_t pm_errno; /** \addtogroup alpm_interface Interface Functions - * @brief Function to initialize and release libalpm + * @brief Functions to initialize and release libalpm * @{ */ @@ -127,15 +127,11 @@ int alpm_release() return(0); } -/** @} */ -/** \addtogroup alpm_options Library Options - * @brief Functions to set and get libalpm options - * @{ - */ +/** @} */ /** \addtogroup alpm_databases Database Functions - * @brief Frunctions to query and manipulate the database of libalpm + * @brief Functions to query and manipulate the database of libalpm * @{ */ @@ -553,7 +549,7 @@ int alpm_pkg_vercmp(const char *ver1, co } /* internal */ -char *_supported_archs[] = { +static char *_supported_archs[] = { "i586", "i686", "ppc", @@ -589,7 +585,6 @@ char *alpm_pkg_name_hasarch(char *pkgnam return NULL; } - /** @} */ /** \addtogroup alpm_sync Sync Functions @@ -611,6 +606,7 @@ pmlist_t *alpm_db_search(pmdb_t *db) return(_alpm_db_search(db, handle->needles)); } + /** @} */ /** \addtogroup alpm_trans Transaction Functions @@ -764,24 +760,7 @@ int alpm_trans_release() return(0); } -/** @} */ -/** \addtogroup alpm_dep Dependency Functions - * @brief Functions to get informations about a libalpm dependency - * @{ - */ -/** @} */ - -/** \addtogroup alpm_conflict File Conflicts Functions - * @brief Functions to get informations about a libalpm file conflict - * @{ - */ - -/** Get informations about a file conflict. - * @param conflict database conflict structure - * @param parm name of the info to get - * @return a void* on success (the value), NULL on error - */ /** @} */ /** \addtogroup alpm_log Logging Functions @@ -894,6 +873,7 @@ int alpm_list_count(const pmlist_t *list return(_alpm_list_count(list)); } + /** @} */ /** \addtogroup alpm_misc Miscellaneous Functions @@ -1111,7 +1091,7 @@ int alpm_parse_config(char *file, alpm_c if(!strcmp(key, "SERVER")) { /* add to the list */ if(alpm_db_setserver(db, ptr) != 0) { - /* pm_errno is set by alpm_set_option */ + /* pm_errno is set by alpm_db_setserver */ return(-1); } } else { @@ -1127,6 +1107,6 @@ int alpm_parse_config(char *file, alpm_c return(0); } -/* @} */ +/** @} */ /* vim: set ts=2 sw=2 noet: */ --- pacman-lib.orig/doc/libalpm.3.in 2006-02-01 13:04:01.000000000 -0500 +++ pacman-lib/doc/libalpm.3.in 2007-01-07 23:23:23.000000000 -0500 @@ -1,4 +1,4 @@ -.TH libalpm 3 "29 Jan 2006" "libalpm @PM_VERSION@" "" +.TH libalpm 3 "07 Jan 2007" "libalpm @PM_VERSION@" "" .SH NAME libalpm \- Arch Linux Package Management library .SH SYNOPSIS @@ -6,13 +6,10 @@ For ease of access, the libalpm manual h .nf alpm_databases Database Functions -alpm_dep Dependency Functions -alpm_groups Group Functions alpm_interface Interface Functions alpm_list List Functions alpm_log Logging Functions alpm_misc Miscellaneous Functions -alpm_options Library Options alpm_packages Package Functions alpm_sync Sync Functions alpm_trans Transaction Functions

2 1

[pacman-dev] pacman-optimize updates
by James Rosten 11 Jan '07

11 Jan '07

Ok, after testing it, some changes needed to be made and so they were made. Now it works (at least for me it worked). This is just an idea, but wouldn't it be interesting if we used the stat_busy, stat_done, and stat_fail stuff from /etc/rc.d/functions for the messages in pacman-optimize? Anyway, here is the patch for FS #5009 (woops, just caught another bug.........it wouldn't be able to work after a successful one but it would work after the failure ('cause pacman.new was still haning around.........fixed). It is below as usual. ~ Jamie / yankees26

3 9

[pacman-dev] Pacman's new feature
by Fabiano Furtado Pessoa Coelho 10 Jan '07

10 Jan '07

Hi people, I´v been thinking about new feature on pacman. Why not to show the each package size on the install package list, just befere the download/instalation? Like: <package name> (SIZE) | <package name> (SIZE) | <package name> (SIZE) Thanks in advance... Fusca PS: Sorry about my english

2 1

[pacman-dev] pacman packaging
by Dan McGee 09 Jan '07

09 Jan '07

This was just another thought of mine. Pacman is currently distributed as one package on Archlinux (and I would assume on any distribution that is using it). I was thinking it should maybe be two packages, pacman and pacman-utils, or something similar. The two things can be split fairly cleanly- the first is used for installing packages and general maintenence, while the other is used for building packages, looking at the ABS tree, building a custom repo, etc. This also seems more important with a few other scripts being added recently (re-pacman, repo-add) pacman: pacman (and conf/man files), pacman.static, vercmp, rankmirrors, pacman-optimize pacman-utils: abs (and conf files), makepkg (and conf file), makeworld (TODO still wondering if this is used), gensync, resync, repo-add (TODO should this be renamed to addsync?), re-pacman In addition, pacman-utils could then pull in scripts such as srcpac and namcap that are likely installed by most developers anyway. This isn't so much development related, but I thought this was the best list to bring this up. -Dan

4 7