Bug 937868 – gnome-base/gnome-shell-45.5 fails to build on musl - gnome-shell-45.5/src/main.c:149:14: error: '_DYNAMIC' undeclared (first use in this function); did you mean 'PT_DYNAMIC'?

• Status: UNCONFIRMED
• Product: Gentoo Linux

[Description Dino 2024-08-13 10:12:24 UTC]

When building gnome-shell-45.5 on Gentoo against musl it will fail at line src/gnome-shell.p/main.c.o with following message:

FAILED: src/gnome-shell.p/main.c.o 
../gnome-shell-45.5/src/main.c: In function 'maybe_add_rpath_introspection_paths':
../gnome-shell-45.5/src/main.c:149:14: error: '_DYNAMIC' undeclared (first use in this function); did you mean 'PT_DYNAMIC'?
  149 |   for (dyn = _DYNAMIC; dyn->d_tag != DT_NULL; dyn++)
../gnome-shell-45.5/src/main.c:149:14: note: each undeclared identifier is reported only once for each function it appears in
../gnome-shell-45.5/src/main.c:162:7: warning: implicit declaration of function 'dladdr1'; did you mean 'dladdr'? [-Wimplicit-function-declaration]
  162 |   if (dladdr1 (_DYNAMIC, &dl_info, (void **) &link_map, RTLD_DL_LINKMAP))
      |       ^~~~~~~
      |       dladdr
../gnome-shell-45.5/src/main.c:162:57: error: 'RTLD_DL_LINKMAP' undeclared (first use in this function); did you mean 'RTLD_DI_LINKMAP'?
  162 |   if (dladdr1 (_DYNAMIC, &dl_info, (void **) &link_map, RTLD_DL_LINKMAP))

During an investigation it seems that the following patch needs to be applied (also ChimeraOS and Void Linux provide this patch on musl): https://github.com/void-linux/void-packages/blob/master/srcpkgs/gnome-shell/patches/no-exe-introspection.patch
After applying this patch, build finishes successfully.

emerge --info log: https://bpa.st/BWJA
gnome-shell-45.5 build log: https://bpa.st/5VSA

flags: 

COMMON_FLAGS="-O2 -march=alderlake -pipe"
CFLAGS="${COMMON_FLAGS}"
CXXFLAGS="${COMMON_FLAGS}"
FCFLAGS="${COMMON_FLAGS}"
FFLAGS="${COMMON_FLAGS}"

CHOST="x86_64-pc-linux-musl

[Comment 1 Dino 2024-08-13 10:14:58 UTC]

Created attachment 900051 [details, diff]
Proposed patch

[Comment 2 Dino 2024-08-13 10:15:40 UTC]

Created attachment 900052 [details]
emerge --info log

[Comment 3 Dino 2024-08-13 10:18:48 UTC]

Created attachment 900053 [details]
gnome-shell-45.5 build log

[Comment 4 immolo 2024-08-13 10:53:27 UTC]

Thanks for the report and testing of this patch.

Would you also be able to test if gnome-base/gnome-shell-45.6-r1 works with and without this patch please?

[Comment 5 Dino 2024-08-13 11:06:55 UTC]

Created attachment 900056 [details]
gnome-shell-45.6-r1 with patch

[Comment 6 Dino 2024-08-13 11:07:19 UTC]

Created attachment 900057 [details]
gnome-shell-45.6-r1 without patch

[Comment 7 Dino 2024-08-13 11:08:54 UTC]

Thanks for your comment. Tested also with gnome-shell-45.6-r1 with and without patch. Produces exactly the same result. With patch - source compiled, without patch - same error. Logs attached.

[Comment 8 Dino 2024-08-13 11:09:40 UTC]

Command used: 

ebuild /var/db/repos/gentoo/gnome-base/gnome-shell/gnome-shell-45.6-r1.ebuild clean compile

[Comment 9 Sven E. 2024-10-08 19:11:51 UTC]

I can at least second the build failure for now.
As far as I can tell, the patch (in a very crude manner) simply disables the feature.

IMHO it would be much better, if the build system would not set HAVE_EXE_INTROSPECTION in the first place.

[Comment 10 Sven E. 2024-10-08 20:52:47 UTC]

For whatever reason I cannot attach with pybugz and since the system with musl is headless, I will do it this way:

diff -Naur gnome-shell-45.6-orig/meson.build gnome-shell-45.6-new/meson.build
--- gnome-shell-45.6-orig/meson.build   2024-04-21 12:32:39.000000000 +0000
+++ gnome-shell-45.6-new/meson.build    2024-10-08 20:06:17.844381514 +0000
@@ -160,7 +160,7 @@
 cdata.set('HAVE_MALLINFO2', cc.has_function('mallinfo2'))
 cdata.set('HAVE_SYS_RESOURCE_H', cc.has_header('sys/resource.h'))
 cdata.set('HAVE_EXE_INTROSPECTION',
-  cc.has_header('elf.h') and cc.has_header('link.h'))
+  cc.has_header('elf.h') and cc.has_header('link.h') and cc.has_header_symbol('link.h','_DYNAMIC'))
 cdata.set('HAVE__NL_TIME_FIRST_WEEKDAY',
   cc.has_header_symbol('langinfo.h', '_NL_TIME_FIRST_WEEKDAY')
 )

This checks link.h for _DYNAMIC which musl does not provide. Fixing the _DYNAMIC issue is trivial, however, musl also does not seem to provide for dladdr1() which is used aswell.