Date: Sat, 02 Oct 2021 11:09:42 +0000 From: bugzilla-noreply@freebsd.org To: x11@FreeBSD.org Subject: [Bug 258463] x11/libwacom: Fix build with clang/lld 13 Message-ID: <bug-258463-7141-XfpNCDkbfu@https.bugs.freebsd.org/bugzilla/> In-Reply-To: <bug-258463-7141@https.bugs.freebsd.org/bugzilla/> References: <bug-258463-7141@https.bugs.freebsd.org/bugzilla/>
next in thread | previous in thread | raw e-mail | index | archive | help
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D258463 --- Comment #6 from commit-hook@FreeBSD.org --- A commit in branch main references this bug: URL: https://cgit.FreeBSD.org/ports/commit/?id=3Da1edd535d35d03e2d54a49bffac3903= 5b990a716 commit a1edd535d35d03e2d54a49bffac39035b990a716 Author: Dimitry Andric <dim@FreeBSD.org> AuthorDate: 2021-09-12 19:44:36 +0000 Commit: Dimitry Andric <dim@FreeBSD.org> CommitDate: 2021-10-02 10:58:25 +0000 x11/libwacom: work around lld 13 no longer supporting symver tricks Though x11/libwacom was not yet built during the exp-run for clang/llvm 13 (see bug 258209), due to some other dependencies not being available yet, I noticed that it failed to build with clang 13, or more specifically this is due to a behavior change in lld 13: ... [ 33% 10/30] cc -o generate-hwdb generate-hwdb.p/tools_generate-hwdb= .c.o -Wl,--as-needed -Wl,--no-undefined -fstack-protector-strong -O2 -pipe -g -fstack-protector-strong -fno-strict-aliasing '-Wl,-rpath,$ORIGIN/' -Wl,-rpath-link,/wrkdirs/share/dim/ports/x11/libwacom/work/libwacom-1.5/_bu= ild/ -Wl,--start-group libwacom.so.2.6.1 /usr/local/lib/libglib-2.0.so /usr/local/lib/libintl.so -Wl,--end-group [ 36% 11/30] /usr/local/bin/meson --internal exe --capture 65-libwacom.hwdb -- /wrkdirs/share/dim/ports/x11/libwacom/work/libwacom-1.5/_build/generate-hwdb FAILED: 65-libwacom.hwdb /usr/local/bin/meson --internal exe --capture 65-libwacom.hwdb -- /wrkdirs/share/dim/ports/x11/libwacom/work/libwacom-1.5/_build/generate-hwdb --- stderr --- Unfortunately the meson build process doesn't really show you why it failed, but it turns out that running the 'generate-hwdb' command segfaults: Starting program: /wrkdirs/share/dim/ports/x11/libwacom/work/libwacom-1.5/_build/generate-hwdb Program received signal SIGSEGV, Segmentation fault. libwacom_add_match (device=3Ddevice@entry=3D0x801031320, newmatch=3D0= x0) at ../libwacom/libwacom.c:943 943 device->matches[device->nmatches - 1] =3D libwacom_match_ref(newmatch); (gdb) bt #0 libwacom_add_match (device=3Ddevice@entry=3D0x801031320, newmatch= =3D0x0) at ../libwacom/libwacom.c:943 #1 0x000000080024fc7d in libwacom_matchstr_to_match (device=3Ddevice@entry=3D0x801031320, matchstr=3D<optimized out>) at ../libwacom/libwacom-database.c:207 #2 0x000000080024e313 in libwacom_parse_tablet_keyfile (db=3D0x80103= 65a0, datadir=3D0x200b70 "/wrkdirs/share/dim/ports/x11/libwacom/work/libwacom-1.5/data", filename=3D<optimized out>) at ../libwacom/libwacom-database.c:652 #3 load_tablet_files (db=3D0x8010365a0, datadir=3D0x200b70 "/wrkdirs/share/dim/ports/x11/libwacom/work/libwacom-1.5/data") at ../libwacom/libwacom-database.c:865 #4 libwacom_database_new_for_path (datadir=3D0x200b70 "/wrkdirs/share/dim/ports/x11/libwacom/work/libwacom-1.5/data") at ../libwacom/libwacom-database.c:959 #5 0x00000000002021b6 in main (argc=3D<optimized out>, argv=3D0x8010= 36630) at ../tools/generate-hwdb.c:131 What happens is that an internal function 'libwacom_match_new' is supposed to be called, which returns a new 'WacomMatch' object. But instead, it calls a empty stub which returns NULL, resulting in this segfault. The empty stub was added as a rather nasty upstream hack to "Alias the accidentally exposed ABI into different functions", in =20=20=20 https://github.com/linuxwacom/libwacom/commit/b9961dbe912fa096230460b194eeb= dc8a590d256: > A special "trick" is used here to hide the ABI from new versions: > Usually when defining multiple versioned symbols, one would define on= e as the > default one with @@ > .symver _foo1,foo@VERSION1 > .symver _foo2,foo@@version2 <-- default one > By leaving out the default one, ld doesn't know which one to link to = and > fails with an unresolved symbol. rtld however can still figure it out= , so > anything compiled will continue to work. This way we can make a symbol > disappear from the library for new builds but have old builds continu= e to > work with the new version. Unfortunately this trick/hack does not work anymore with lld 13, since https://github.com/llvm/llvm-project/commit/66d44304921, ("[ELF] Combine foo@v1 and foo with the same versionId if both are defined "). The idea behind the hack is to have the linker call the 'real' libwacom_match_new function whenever it is called from inside the library itself, but any external callers get the stubbed version which doesn't really do anything. I think libwacom should have used a different approach here, but just renaming those accidentally exposed internal functions to something different. Then the tricks with .symver are completely unnecessary. Here I added a patch that is as simple as possible, which adds #defines for two affected functions in libwacomint.h, renaming then from 'libwacom_xxx' to 'libwacom_internal_xxx'. This does not affect the corresponding exposed functions in the libwacom.so, and makes the 'generate-hwdb' command work OK again. I also ran the complete libwacom test suite, including the deprecated functions test, and it works fine. PR: 258463 Approved by: zeising (maintainer) MFH: 2021Q4 x11/libwacom/files/patch-libwacom_libwacomint.h (new) | 12 ++++++++++++ 1 file changed, 12 insertions(+) --=20 You are receiving this mail because: You are the assignee for the bug.=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-258463-7141-XfpNCDkbfu>