Bug 128596 – 'ar T' is not portable and breaks the build on FreeBSD

• Component: WebKit Misc.
• Status: RESOLVED FIXED

[Allison Lortie (desrt)]

Source/autotools/SetupLibtool.m4 has:

if test -z "$AR_FLAGS"; then
    AR_FLAGS="cruT"
fi
AC_SUBST([AR_FLAGS])


GNU ar documents the 'T' flag like so:

       T   Make the specified archive a thin archive.  If it already exists and is a regular archive, the existing
           members must be present in the same directory as archive.

FreeBSD ar documents the 'T' flag like so:

     -T      Use only the first fifteen characters of the archive member name
             or command line file name argument when naming archive members.


This fragment of Source/WebKit2/GNUmakefile.am:

libWebCoreLayerGtk2.a: $(webcore_layer_gtk2_deps)
        $(AM_V_GEN)
        $(AM_V_at)$(shell rm -f $@)
        $(AM_V_at)$(shell find . -name "*.o" > objects_list)
        $(AM_V_at)$(foreach archive, $(webcore_layer_gtk2_archives), $(shell $(AR) t $(archive) | xargs -n1 basename | xargs -I obj_file grep -F obj_file objects_list | xargs -n50 $(AR) $(AR_FLAGS) $@))
        $(AM_V_at)$(shell rm -f objects_list)


results in a bunch of arguments like './Source/WebCore/platform/graphics/libPlatform_la-FloatRect.o' being passed to 'ar'.

Of course, most of those arguments will share the same first 15 characters, and as a result, a large number of files are not included in the archive.



Webkit should not use 'ar T' on systems with a non-GNU toolchain, or it should explicitly require a GNU toolchain and make sure it uses that.

[Allison Lortie (desrt)]

Created attachment 223846 [details]
SetupLibtool.m4: use 'ar T' only on GNU ar

[Alberto Garcia]

Created attachment 223848 [details]
Patch

The patch looks fine, here's the updated one with the changelog entry.

Thanks!

[Alberto Garcia]

Committed r163954: <http://trac.webkit.org/changeset/163954>

[Landry Breuil]

I know this has been fixed/commited, but as it was already sort-of fixed in https://bugs.webkit.org/show_bug.cgi?id=118732, couldnt you set AR_FLAGS=cru in the build environment to avoid having to patch autohell this way ?

[Allison Lortie (desrt)]

I don't think it should be expected to have to set an environment variable to prevent it from containing non-POSIX values...

If anything, it should be the other way around.

I think having this check can make everyone happy.

[Landry Breuil]

(In reply to comment #5)
> I don't think it should be expected to have to set an environment variable to prevent it from containing non-POSIX values...
> 
> If anything, it should be the other way around.
> 
> I think having this check can make everyone happy.

I totally agree (and i'm happy with the check for OpenBSD too!), was just wondering about the reasoning - usually it's not so easy to get non-linux fixes in :)

[Alberto Garcia]

(In reply to comment #6)
> I totally agree (and i'm happy with the check for OpenBSD too!), was
> just wondering about the reasoning - usually it's not so easy to get
> non-linux fixes in :)

The solution proposed by Ryan looked just fine to me. And I've been
pushing non-linux fixes for a long time btw :)