[Pkg-pulseaudio-devel] Bug#576546: Feedback

[Marcelo E. Magallon]

Hi,

 I'm not sure how this ever worked.  With the shipped
 /usr/lib/libpulsecommon-0.9.21.so pa_make_secure_dir
 consistenly fails.

 As someone else pointed out the problem is here:

        (st.st_uid != uid) ||
        (st.st_gid != gid) ||

 because at this point uid = -1 and gid = -1.

 Looking at the build log:

 https://buildd.debian.org/fetch.cgi?&pkg=pulseaudio&ver=0.9.21-1.1&arch=amd64&stamp=1270339860&file=log

 it looks like fstat and fchown were correctly detected,
 presumably defining HAVE_FSTAT and HAVE_FCHOWN:

#ifdef HAVE_FSTAT
...

#ifdef HAVE_FCHOWN
    if (uid == (uid_t)-1)
        uid = getuid();
    if (gid == (gid_t)-1)
        gid = getgid();
    (void) fchown(fd, uid, gid);
#endif

...
#endif

 I rebuilt the package locally and it's working (in my working
 environment, not a chroot), so it looks like there's a missing
 build-dependency, my guess is autotools *and* a call to
 autoconf.

 Examining this more closely, doing a clean build of the source
 package as uploaded results in a binary with this bug.  When
 doing debian/rules clean, I see autoconf being called, which
 modifies the source and on the second build the package works
 correctly.  So something in CDBS is failing to call autoconf
 even if configure.ac is being patched.

 Two bugs were reported about the same issue, one on amd64 and
 one on i386.  Looking at the autobuilt set, it seems like the
 original NMU was built on i386, so I don't think this problem is
 specific to the autobuilt versions.

 Also, I think the patch breaks pulseaudio somewhere, because it
 hides this code:

    if (uid == (uid_t)-1)
        uid = getuid();
    if (gid == (gid_t)-1)
        gid = getgid();

 inside the two HAVE_FCHOWN and HAVE_FSTAT conditionals.  The
 original code had it hidden inside a single conditional on
 HAVE_CHOWN.  My point is I don't think the patched code is doing
 the same wrt the uid/gid variables.

 Looking at the structure of the code, I think those lines should
 be outside the conditional, to support the use case where the
 caller of pa_make_secure_dir passes -1 as uid/gid, but there's
 no support for fstat/fchown/chown, or else the check further
 down the code fails consistently there.

 So, there are two problems: for whatever reason HAVE_FSTAT
 and/or HAVE_FCHOWN weren't defined; and the patch is breaking
 pulseaudio somewhere (likely not Linux).

 Marcelo