Perform the cheap struct and string setup of the local DB at handle
initialization time to match the teardown we do when releasing the handle.
If the local DB is not needed, all real initialization is done lazily after
DB paths and other things have been configured anyway.
Signed-off-by: Dan McGee <dan@archlinux.org>
We explicitly place 'pkgbase' (and used to place 'force') fields inside
PKGINFO files, so ignore them silently instead of printing an error for
them. Also make the error message for unknown keys actually contain the key.
Signed-off-by: Dan McGee <dan@archlinux.org>
We were returning a package error code rather than a DB one, and we
would leak the archive memory if the database file didn't exist.
Signed-off-by: Dan McGee <dan@archlinux.org>
Instead, go the same route we have always taken with version-release in
libalpm and treat it all as one piece of information. Makepkg is the only
script that knows about epoch as a distinct value; from there on out we will
parse out the components as necessary.
This makes the code a lot simpler as far as epoch handling goes. The
downside here is that we are tossing some compatibility to the wind;
packages using force will have to be rebuilt with an incremented epoch to
keep their special status.
Signed-off-by: Dan McGee <dan@archlinux.org>
Adapting from RPM, follow the [epoch:]version[-release] syntax. We can also
borrow some of their parsing code for our purposes (thanks!). Add some new
tests to our vercmp shell script tester for epoch comparisons, and then make
the code work with these newfangled epoch specifiers.
Signed-off-by: Dan McGee <dan@archlinux.org>
Since it is the same string. Done with some bash looping and sed magic.
for src in po/*.po; do
echo $src
newtrans=$(grep -A1 "msgid.*$1" $src | tail -n1)
newtrans=${newtrans//\\/\\\\}
echo "$newtrans"
fname=${src##*/}
dest=lib/libalpm/po/$fname
sed -i -e "/msgid.*$1/{N; s/msgstr.*$/$newtrans/}" $dest
done
Signed-off-by: Dan McGee <dan@archlinux.org>
In most (all?) cases, we will process all files for a given sync database
entry sequentially. The code currently does an _alpm_pkg_find() for every
file in the database, but we had the "current" package readily available.
Shift some local variables around a bit to expose this to sync_db_read() and
use it if the package is the correct one.
On my system, this cuts calls to _alpm_pkg_find() from 20,769 to 10,349
calls during a -Qu operation, and results in a ~30% speedup of the same
operation (0.35 sec -> 0.27 sec). This benefit should be apparent anywhere
we read in the full contents of the sync databases.
Signed-off-by: Dan McGee <dan@archlinux.org>
We don't need to create a temporary copy of the string if we are smart with
our pointer manipulation and string copying. This saves a bunch of string
duplication during database parsing, both local and sync.
Signed-off-by: Dan McGee <dan@archlinux.org>
Remove the need for an unconditional string duplication by using pointer
arithmetic instead, and strndup() instead of an unspecified-length strdup().
This should reduce memory churn a fair amount as this is called pretty
frequently during database loads.
Signed-off-by: Dan McGee <dan@archlinux.org>
When installing packages from a file, the integrity check count
stays at (0/x) complete. This ensures it is bumped to (x/x) at
the end of the process.
Signed-off-by: Dan McGee <dan@archlinux.org>
This is essentially a backport/cherry-pick of commit 33240e87b9 from
master, but has to be done by hand because the DB format has diverged. Read
more in the commit message used there, which follows.
Due to the way we funk around with package data loading, we had a condition
where the filelist got doubled up because it was loaded twice.
Packages are originally loaded with INFRQ_BASE. In an upgrade/sync, the
package is checked for file conflicts next, leaving us in an "INFRQ_BASE |
INFRQ_FILES" state. Later, when committing a single package, we have an
explicit call to _alpm_local_db_read() with INFRQ_ALL as the level. Because
the package's level did not match this, we skipped over our previous "does
the incoming level match where I'm at" shortcut, and continued to load
things again, because of a lack of fine-grained checking for each of DESC,
FILES, and INSTALL.
The end result is we loaded the filelist twice, causing our remove logic to
iterate twice over the installed files, spewing a bunch of "cannot find file
X" messages.
Fix the problem by doing a bit more bitmasking logic throughout the load
method, and also fix the sanity check at the beginning of the function- this
should *only* be used for local packages as opposed to the "not a package"
check that was there before.
Signed-off-by: Dan McGee <dan@archlinux.org>
After all the debate as to what to do on maint, we are going to end up just
incorporating epoch into the version string, so we don't need this separate
field at all. Revert commit 5c8083baa4 and also kill the force flag we were
recording here as well.
Signed-off-by: Dan McGee <dan@archlinux.org>
Due to the way we funk around with package data loading, we had a condition
where the filelist got doubled up because it was loaded twice.
Packages are originally loaded with INFRQ_BASE. In an upgrade/sync, the
package is checked for file conflicts next, leaving us in an "INFRQ_BASE |
INFRQ_FILES" state. Later, when committing a single package, we have an
explicit call to _alpm_local_db_read() with INFRQ_ALL as the level. Because
the package's level did not match this, we skipped over our previous "does
the incoming level match where I'm at" shortcut, and continued to load
things again, because of a lack of fine-grained checking for each of DESC,
FILES, and INSTALL.
The end result is we loaded the filelist twice, causing our remove logic to
iterate twice over the installed files, spewing a bunch of "cannot find file
X" messages.
Fix the problem by doing a bit more bitmasking logic throughout the load
method, and also fix the sanity check at the beginning of the function- this
should *only* be used for local packages as opposed to the "not a package"
check that was there before.
A debug log message was added to upgraderemove as well to match the one
already in the normal remove codepath.
Signed-off-by: Dan McGee <dan@archlinux.org>
libarchive eventually calls it anyway, but backtraces make a lot more sense
if we call it, as well as matching our precedent from alpm_pkg_load().
Signed-off-by: Dan McGee <dan@archlinux.org>
This is helpful anyway to the user, and should also be helpful to us if we
see problems cropping up in the check during development.
Also add a missing ->used = 0 initialization in the code path less taken.
Signed-off-by: Dan McGee <dan@archlinux.org>
Although they won't be the same in the gettext catalog because of the '\n'
we should still use the same text.
Signed-off-by: Dan McGee <dan@archlinux.org>
This can take a while too, and it is really easy to add the necessary
callback stuff for adding a progressbar.
Signed-off-by: Dan McGee <dan@archlinux.org>
These were just two small things I came across today and found could be
fixed or helpful, so I've added them and I'm not sure what else to bundle
them with. commit_count++
Signed-off-by: Dan McGee <dan@archlinux.org>
We were checking if a package existed locally, but then using the
incoming package to calculate removed size rather than the currently
installed package.
Also adjust the local variable in the replaces loop to make it more
clear that we are always dealing with local packages here.
Signed-off-by: Dan McGee <dan@archlinux.org>
All of these can be done with integer division; the only slightly
interesting part is ensuring we round up like before with calling the
ceil() function.
We can also remove the math library from requirements; now that the only
ceil() calls are gone, we don't need this anymore.
Signed-off-by: Dan McGee <dan@archlinux.org>
file_pkg_ops can be a static struct like in other backends, we just need
to initialize it at some point.
Dan: add initialization flag.
Signed-off-by: Xavier Chantry <chantry.xavier@gmail.com>
Signed-off-by: Dan McGee <dan@archlinux.org>
None of these warn at the normal "-Wall -Werror" level, but casts do occur
that we are fine with. Make them explicit to silence some warnings when
using "-Wconversion".
Signed-off-by: Dan McGee <dan@archlinux.org>
There is a lot of swtiching between size_t and int for alpm_list sizes
in the codebase. Start converting these to all be size_t by adjusting
the return type of alpm_list_count and fixing all additional warnings
given by -Wconversion that are generated by this change.
Dan: a few more small changes to ensure things compile, adjusting some
printf format string characters to accommodate the larger size on x86_64.
Signed-off-by: Allan McRae <allan@archlinux.org>
Signed-off-by: Dan McGee <dan@archlinux.org>
All functions that are limited to the local translation unit are
declared static. This exposed that the _pkg_get_deltas declaration
in be_local.c was being satified by the function in packages.c which
when declared static caused linker failures.
Fixes all warnings with -Wmissing-{declarations,prototypes}.
Signed-off-by: Allan McRae <allan@archlinux.org>
Signed-off-by: Dan McGee <dan@archlinux.org>
We use PATH_MAX everywhere by including limits.h so there is no
point in doing a check for it in a different header when dealing
with FreeBSD's libfetch.
Also, remove autoconf check for strings.h header as it is not used
anywhere.
Signed-off-by: Allan McRae <allan@archlinux.org>
Signed-off-by: Dan McGee <dan@archlinux.org>
POSIX does not require PATH_MAX be defined when there is not actual
limit to its value. This affects HURD based systems. Work around
this by defining PATH_MAX to 4096 (as on Linux) when this is not
defined.
Also, clean up inclusions of limits.h and remove autoconf check for
this header as we do not use macro shields for its inclusion anyway.
Signed-off-by: Allan McRae <allan@archlinux.org>
Signed-off-by: Dan McGee <dan@archlinux.org>
Just like we did for package name comparsions, if we add a depend name_hash
field on depend struct initialization, we can use it instead of doing a
string name comparison, saving us a lot of checks in the depcmp code.
Signed-off-by: Dan McGee <dan@archlinux.org>
Noticed when tweaking testdb, when we run _alpm_depcmp in loops and call it
seven million times, the strdup()/free() combo can add up. Remove the need
for any string duplication by some pointer manipulation and use of strncmp
instead of strcmp. Also kill the function logger and add an escape so we
don't needlessly retrieve the list of provides.
Signed-off-by: Dan McGee <dan@archlinux.org>
The old function was written in a time before we relied on it for nearly
every operation. Since then, we have switched to the archive backend and now
fast parsing is a big deal.
The former function made a per-character call to the libarchive
archive_read_data() function, which resulted in some 21 million calls in a
typical "load all sync dbs" operation. If we instead do some buffering of
our own and read the blocks directly, and then find our newlines from there,
we can cut out the multiple layers of overhead and go from archive to parsed
data much quicker.
Both users of the former function are switched over to the new signature,
made easier by the macros now in place in the sync backend parsing code.
Performance: for a `pacman -Su` (no upgrades available),
_alpm_archive_fgets() goes from being 29% of the total time to 12% The time
spent on the libarchive function being called dropped from 24% to 6%.
This pushes _alpm_pkg_find back to the title of slowest low-level function.
Signed-off-by: Dan McGee <dan@archlinux.org>
The amount of diskspace needed for a transaction can be less than
zero. Only test this against the available disk space if it is
positive, which avoids a comparison being made between signed and
unsigned types (-Wsign-compare).
Signed-off-by: Allan McRae <allan@archlinux.org>
Signed-off-by: Dan McGee <dan@archlinux.org>
Always declare a function with (void) rather than () when we expect
no arguements. Fixes all warnings with -Wstrict-prototypes.
Signed-off-by: Allan McRae <allan@archlinux.org>
Signed-off-by: Dan McGee <dan@archlinux.org>
This simplifies a lot of the repetative code and makes it obvious where the
tricky or different ones are (e.g. depends, dates). It also makes it
significantly easier to change the way this code works in the future.
There should be no functional change with this patch.
Signed-off-by: Dan McGee <dan@archlinux.org>
Saves a few bytes due to padding (256 -> 248 bytes), especially on x86_64,
so we get the overhead of our new hash field right back.
Signed-off-by: Dan McGee <dan@archlinux.org>
This results in huge gains to a lot of our codepaths since this is the most
frequent method of random access to packages in a list. The gains are seen
in both profiling and real life.
$ pacman -Sii zvbi
real: 0.41 sec -> 0.32 sec
strcmp: 16,669,760 calls -> 473,942 calls
_alpm_pkg_find: 52.73% -> 26.31% of time
$ pacman -Su (no upgrades found)
real: 0.40 sec -> 0.50 sec
strcmp: 19,497,226 calls -> 524,097 calls
_alpm_pkg_find: 52.36% -> 26.15% of time
There is some minor risk with this patch, but most of it should be avoided
by falling back to strcmp() if we encounter a package with a '0' hash value
(which we should not via any existing code path). We also do a strcmp once
hash values match to ensure against hash collisions. The risk left is that a
package name is modified once it was originally set, but the hash value is
left alone. That would probably result in a lot of other problems anyway.
Signed-off-by: Dan McGee <dan@archlinux.org>
