- Pacman-dev - lists.archlinux.org

[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 10 Jan '07

10 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

[pacman-dev] Install files and messages
by Dan McGee 10 Jan '07

10 Jan '07

I don't know what other peoples thoughts are, but I think the current install file system and scriptlets...seems like a drop in at some point that wasn't fully done right. Most packages currently use the file for one of two purposes that I can see: 1. display informational messages 2. perform operations that can't be done in the build phase. The install scriptlets were likely originally meant more for number 2, but has become more and more about providing number 1. Here are some relevant bugs and postings: metadata messages in PKGBUILD <http://bugs.archlinux.org/task/1571> display all post-install messages at once <http://bugs.archlinux.org/task/6143> global on-install system <http://bbs.archlinux.org/viewtopic.php?p=219152#219152> Here are my thoughts on the situation. As every package builder seems to have their own presentation format for displaying messages (e.g. echo ">>>...", echo " **...**"), it would be nice to have a way to unify it. Most (all?) messages are in one of two places- post_install or post_upgrade. I propose these messages should be moved out of these scriptlets, and moved into a variable, likely still located in the install file as this is packaged up with the package (unlike the PKGBUILD). These variables could be called install_msg and upgrade_msg, respectively. The scriptlets would still remain, but any output would be 'nonessential'. By changing where the messages stem from, it would make it easier for bug 6143 to be fixed, and would also make for cleaner output to pacman.log, now that all messages are being logged there. Finally, _alpm_runscriptlet seems pretty complicated...I'm wondering if it couldn't be cleaned up a bit by someone that is better with forks and popens. There is also some duplicate code. -Dan

4 5

[pacman-dev] makepkg fix (FS #5923)
by James Rosten 06 Jan '07

06 Jan '07

This is a fix for Flyspray bug #5923 http://bugs.archlinux.org/task/5923. Patch is below. ~ Jamie / yankees26

3 2

[pacman-dev] Long functions and other minor fixes
by James Rosten 03 Jan '07

03 Jan '07

This is my massacre of _alpm_load_pkg in package.c (just joking about the massacre). I moved the extremely large for loop into its own function (and made it a while loop). Feel free to change the name of _alpm_pkg_checker (first thing that came to mind). The patches for package.c and package.h are below. ~ Jamie / yankees26

3 4