Since the tests are only useful either if you're hacking at pulseaudio as
a developer, or when running "make check", allow users to opt-out from
their build.
This for instance allows for Gentoo users not to build the tests when
installing the ebuild with tests disabled, and also allow for skipping over
eventually broken tests when trying to get the basic build going on a port.
With this change, instead of running multiple `mkdir -p` commands in all
the rules for all the generated definition files, only the call for the
current generated file is executed.
Not only it should shorten build time (especially for parallel make) but it
also fixes out-of-tree builds when new directories are added.
The $(dir $@) construct is GNU make-specific, but the rest of the
buildsystem is already GNU make-dependent so there should be no problem.
Grr.. You can't do:
SBC_FILES = a b c
libsbc_SOURCES = $(addprefix modules/bluetooth/,$(SBC_FILES))
With automake...
So I $(strip 'modules/bluetooth/') instead.
Run simple core tests during 'check'.
I plan to add more automated tests, including pulseaudio instances. I
did some for maemo, they still need some love.
Note: it would be funky to enable Automake 'color-tests' option when
this will be released (1.10.2 and from automake git)
Don't mess up with autofoo, to pass 'make distcheck' again.
Note: I am not convinced by the 'rm -f *.la', I know it can solve
issue, but it should be handled by distributions probably...
I guess it used to be builddir because some of the files under
src/modules are generated. However, they are generated at dist time,
not at compile time.
According to Daniel, AC_PATH_XTRA is sort-of deprecated.
This patch changes the configure arguments, --x-includes=dir and
--x-libraries=dir, which are now removed and use standard pkg-config.
It also replaces --with{,out}-x with --{dis,en}able-x11, which is the
same as other optionnal dependencies.
(this patch was done in an attempt to solve a build issue on maemo, it
turns out it didn't help)
When an .so is unloaded during runtime all TLS keys it has registered
need to be freed because the destructor callbacks of the TLS key might
otherwise point to invalid code. Hence it would appear sensible to
destruct the TLS keys from a function marked as __attribute__
((destructor)). However functions marked like that are also called when
an application terminates, on exit(). If a thread continues to run until
the very exit it might still want to access that TLS data. The
destructor functions are called while all other threads are still
running. If __attribute ((destructor)) is used to destruct TLS keys for
such threads this might hence cause a crash when the application shuts
down.
To circumvent this problem we'll now compile libpulse with -z nodelete,
to make it unnecessary to delete the TLS data ever and thus avoiding the
problem. It's suboptimal, but for now I see no better solution.
This also does some minor reordering in the Makefile.am
Refs #69
git-svn-id: file:///home/lennart/svn/public/pulseaudio/branches/coling@2519 fefdeb5f-60dc-0310-8127-8f9354f1896f
This currently hacks in -lssl rather than writing a configure hook to detect it as I want to replace this with nss before official release.
git-svn-id: file:///home/lennart/svn/public/pulseaudio/branches/coling@2397 fefdeb5f-60dc-0310-8127-8f9354f1896f
It still requires a socket client and callback system to be added before it will be functional.
git-svn-id: file:///home/lennart/svn/public/pulseaudio/branches/coling@2366 fefdeb5f-60dc-0310-8127-8f9354f1896f
