MSYS mangles UNIX paths on execution such that it prepends its root in
Windows format. This affects PKG_CONFIG_SYSROOT_DIR and thus the
check-sysroot test. Detect MSYS from the OSTYPE environment variable and
adjust the expected test results to match. This likely means that
sysroot support doesn't actually work on MSYS, but there probably aren't
any toolchains that support sysroot there, anyway.
When printing warnings on stderr that don't immediately exit pkg-config,
flush it so that the messages appear in order with stdout. This is
mostly to keep the test suite passing on Windows where output may appear
differently than on Linux.
When check-path is run on native Windows (MINGW), that path reported
by `pkg-config --variable=pc_path pkg-config` is adjusted to the
directory pkg-config is run from. Adjust the expected result based on
getting the current directory from cmd in Windows format.
The host triplet describes the machine the program will run on while the
build triplet describes the machine the program is being built on. The
determination of native Win32 should be based on where pkg-config is
going to be running. This makes pkg-config cross-compiled for windows
work as if it was built on windows.
By using our PKG_CHECK_MODULES from our in-tree pkg.m4, the check for
glib will be done the same way pkg-config is used everywhere else. This
includes the usage of AC_PATH_TOOL in PKG_PROG_PKG_CONFIG, which will
check for $host-pkg-config when --host is set during configure.
Freedesktop #59435 (https://bugs.freedesktop.org/show_bug.cgi?id=59435)
Instead of keeping of list of packages in the current Requires chain and
searching it repeatedly, just mark each package as part of the chain or
not. This nearly cuts in half the time on a particularly rough torture
test I have.
When the basename of the libdir is lib64, we currently add the paths
/usr/lib64:/lib64 to the generic system library path of /usr/lib:/lib.
Extend this coverage to other valid ABIs that use libdirs such as lib32
or libx32.
The intention was that errors would be printed for all output options
besides --exists and --atleast/exact/max-version, which are intended to
operate silently. Since want_exists is always set for these latter
options, we can simply use that as the condition and catch all other
output options automatically.
Freedesktop #54390 (https://bugs.freedesktop.org/show_bug.cgi?id=54390)
This provides the user with output matching the behavior of the code.
When multiple --atleast/exact/max-version options are supplied, only the
first will be honored.
The --atleast/exact/max-version help description implied that it would
return as --exists does. However, this would only occur if no other
output options were set.
Freedesktop #54389 (https://bugs.freedesktop.org/show_bug.cgi?id=54389)
This happened basically by accident before when "pkg-config foo" was run
because the code wouldn't find any options set and just fall through to
the end after processing the package arguments. However, it would act
differently in that Requires.private was only enabled with an explicit
--exists.
Currently, any output option (e.g., --version or --libs) will be set as
valid and what's output is at the mercy of the order of the output
handling code in main(). However, most combinations of output would make
no sense to be used together. For example, mixing --modversion and
--print-provides provides no way to differentiate between the output
from the options. Further, mixing --variable and --cflags currently
causes an error because there's no space separating the option outputs.
Instead, keep track of when an output option has been set and ignore
subsequent output options. There are currently two exceptions:
1. Any combination of --cflags* and --libs* are allowed.
2. Both --print-requires and --print-requires-private can be used
together as the user may just not care which is private.
Freedesktop #54391 (https://bugs.freedesktop.org/show_bug.cgi?id=54391)
With the output options gathered in a callback, we can be more clever
with the --cflags/--libs options and set the flags mask straight off
without using the intermediate booleans.
The code handling processing of the packages from the command line was
in an awkward block in main. Split it out to a separate function to keep
main mostly about output.
Having the option entries within main and requiring the option variables
to be static is just wrong. Just declare them at file scope like every
other program does.
pkg-config strips all duplicate arguments from the flag output string.
This is done for 2 reasons:
1. When a package shows up twice in the final package list after
resolving all Requires, stripping was used to ensure it's flags only
showed up once at the correct location.
2. An optimization so that the output string is not excessively long.
Since commit c6ec7869, 1. is no longer necessary as the final package
list only contains each package once. 2. causes problems when applied
too aggressively since some arguments have different semantics depending
on the prior or subsequent arguments.
To keep a bit of optimization, the stripping is reduced to only removing
consecutive duplicate arguments. This should ensure that the semantics
are kept intact while removing obviously unnecessary arguments.
The drawback is that some arguments will now appear multiple times in
the output when they previously would have only appeared once. Here we
have to rely on the tools using these arguments to handle the duplicates
appropriately since there is no way for pkg-config to encode all the
semantics of those arguments. Another thing that can help this situation
is if pkg-config is used for all packages in the Requires chain so that
the Libs/Cflags of each package only pertain to itself and don't encode
the compiling/linking rules of a 3rd party package.
Freedesktop #16101 (https://bugs.freedesktop.org/show_bug.cgi?id=16101)
Often other Libs flags have semantics that are based on their context
with -l arguments. For example, the GNU linker options
-Bdynamic/-Bstatic pertain only to the objects or link options that
follow them. So, a valid link command containing these options would get
mangled by pkg-config since it separates -l flags from others..
-Bdynamic -la -Bstatic -lb -> -Bdynamic -Bstatic -la -lb
Instead, output -l and other Libs flags in a single pass so they mantain
their ordering.
Freedesktop #19950 (https://bugs.freedesktop.org/show_bug.cgi?id=19950)
Instead of splitting to -l/-L/other and -I/other, keep the args together
and mark each argument with its type. Then we can maintain order all the
way through.
Outputting other Libs flags such as -Wl,foo just prior to the -l Libs
flags gives a better chance the --libs output will be correct. This
should be no change in the usage of the output since pkg-config
currently groups all flag types together.
Unify the get_multi_merged functions since there are only two valid ways
to do the merging of packages and flags.
1. Packages are sorted by their position in the pkg-config path and then
duplicate flags are stripped from the beginning of the list. This
pertains to -I and -L flags.
2. Packages are sorted by dependency with most required last and then
duplicate flags are stripped from the end of the list. This ensures
that flags that come from packages required by multiple others come
later in the output. This applies to all non-L/I flags.
There were two string list stripping functions, when the only difference
was to operate from the beginning or end of the list. Also, the function
was copying the list and operating on that unnecessarily.
For some flags, pkg-config will sort them by the depth of their .pc path
before outputting. The idea is that flags from a deeper path should come
earlier in the command line. This makes sense for -L and -I flags, but
not for generic linker and compiler flags. For these flags and -l flags,
it makes sense to sort them only by package order.
Commit 428335e changed some hash tables to be used as sets where the key
equals the value since glib has some optimization for that.
Unfortunately, the package list stripping table wasn't fixed corectly.
Prior to commit 6ecf318, the resolved list of required packages was
built in an appending way where each package on the command line or in
Requires would appear in the list in the order they appeared. With
6ecf318, that list building was changed to prepending, which had a
subtle change on the resolved order.
For example, suppose package a has "Requires: b c d". Previously, the
list would be built as a->b->c->d by appending each as they were
encountered. Now, the list is built by walking all the way down the
dependency chain for each package in a depth first manner and prepending
packages while unwinding. This would result in the package ilst being
a->d->c->b. This same effect happens with the command line packages
where previously requesting packages x and y would create a package list
of x->y and now produces a list of y->x.
While technically these should be the same since there are no
interdependencies, it's causes flags to be output in different order
than previously in pkg-config. This can be seen most readily in the
check-gtk test.
Instead, operate on the package lists backwards when building the
resolved package list.
Walking a GSList backwards involved copying and reversing it so that the
the original list could remain undisturbed. This is wasteful with a
GList where we can just start at the end of the list and work backwards.
Makes the resolved package list be correctly serialized with each
package only appearing once. This provides more consistency between the
various flag outputs by ensuring that the flags from each package are
only grabbed once. This makes a difference since the duplicate flag
stripping happens from the end of the output (-l) or the beginning of
the output (-L/-I/other).
Using a doubly-linked list allows it to be easily traversed in both
directions and makes removing nodes in place much simpler. This adds an
extra pointer to each node and associated manipulation during any list
processing, but this trade seems acceptable over the repeated hacks to
work with singly-linked lists.
Often the expected results for the indirect dependency tests fell behind
because it's not a typical test scenario. However, since the results are
always the same as --static, they can just use the same results and the
test can be run conditionally without --static based on configuration.
Using the GSList functions instead of manually adjusting the list
pointers seems safer and allows an easier path to using another glib
list type if necessary.
The pkg-config testsuite has pretty good coverage of the implementation,
but it lacks a complex case that tests the interactions of non-trivial
.pc files. gtk is a very common package that meets this goal. This is a
snapshot from my F16 system, and it should provide a good way to see how
changes in the implementation regress a real world case.
pkg-config aggressively strips all duplicate arguments from the final
output it builds. This is not only and optimization, but it also allows
the flag ordering to work correctly when a package on the command line
is required by another on the command line.
The current tests are good at checking whether gathering the Cflags or
Libs from one or two packages works correctly, but they don't check the
sorting algorithm much at all. In particular, the interactions between
the package order in the Requires chain and in the path can make the
sorting of the flags subtly different.
