Bug 899742 – net-libs/xdp-tools-1.3.1-r3 fails to compile: Warning: Num of global symbols in libxdp.o xsk.o (68) does NOT match with num of versioned symbols in libxdp.so (71) Please make sure all symbols are versioned in libxdp.map

• Status: CONFIRMED
• Product: Gentoo Linux

[Description Agostino Sarubbo 2023-03-06 07:58:47 UTC]

https://blogs.gentoo.org/ago/2020/07/04/gentoo-tinderbox/

Issue: net-libs/xdp-tools-1.3.1 fails to compile.
Discovered on: amd64 (internal ref: ci)

[Comment 1 Agostino Sarubbo 2023-03-06 07:58:49 UTC]

Created attachment 856228 [details]
build.log

build log and emerge --info

[Comment 2 Holger Hoffstätte 2023-03-06 09:02:05 UTC]

I cannot recreate this at all, even when using all the same toolchain options, same bpftool version, same binutils, same CFLAGS etc. It .. just works?

...
ln -sf libxdp.so.1 libxdp.so
x86_64-pc-linux-gnu-ar rcs libxdp.a staticobjs/libxdp.o staticobjs/xsk.o xdp-dispatcher.embed.o xsk_def_xdp_prog.embed.o xsk_def_xdp_prog_5.3.embed.o
make[2]: Leaving directory '/tmp/portage/net-libs/xdp-tools-1.3.1/work/xdp-tools-1.3.1/lib/libxdp'
...

Why does the tinderbox log show the warnings? The only difference I have found so far is that my gcc/binutils have zstd support enabled, as they should. But that shouldn't bother ar.

[Comment 3 Agostino Sarubbo 2023-03-06 09:13:38 UTC]

(In reply to Holger Hoffstätte from comment #2)
> I cannot recreate this at all, even when using all the same toolchain
> options, same bpftool version, same binutils, same CFLAGS etc. It .. just
> works?
> 
> ...
> ln -sf libxdp.so.1 libxdp.so
> x86_64-pc-linux-gnu-ar rcs libxdp.a staticobjs/libxdp.o staticobjs/xsk.o
> xdp-dispatcher.embed.o xsk_def_xdp_prog.embed.o xsk_def_xdp_prog_5.3.embed.o
> make[2]: Leaving directory
> '/tmp/portage/net-libs/xdp-tools-1.3.1/work/xdp-tools-1.3.1/lib/libxdp'
> ...
> 
> Why does the tinderbox log show the warnings? The only difference I have
> found so far is that my gcc/binutils have zstd support enabled, as they
> should. But that shouldn't bother ar.

can you try with -native-symlinks ?

[Comment 4 Holger Hoffstätte 2023-03-06 09:24:12 UTC]

See also: https://github.com/xdp-project/xdp-tools/issues/302

[Comment 5 Sam James 2023-03-06 09:36:47 UTC]

No ideas yet, I assumed it was an LLVM build at first (seemed like the kind of thing the LLD 16 changes for stricter linker script stuff) but nope.

[Comment 6 Holger Hoffstätte 2023-03-06 09:38:23 UTC]

(In reply to Agostino Sarubbo from comment #3)
> > Why does the tinderbox log show the warnings? The only difference I have
> > found so far is that my gcc/binutils have zstd support enabled, as they
> > should. But that shouldn't bother ar.
> 
> can you try with -native-symlinks ?

I had it on profile-default (unset) and therefore enabled per-package. I explicitly disabled it globally, rebuilt llvm/clang/lld-toolchain-symlinks and tried again - no change, still no warnings from ar.

[Comment 7 Holger Hoffstätte 2023-03-06 11:28:20 UTC]

I just saw that the tinderbox gcc is built with +lto, which also includes ar/nm/ranlib. Do we know that this works reliably?

[Comment 8 Agostino Sarubbo 2023-03-07 08:04:00 UTC]

ci has reproduced this issue with version 1.3.1-r1 - Updating summary.

[Comment 9 Agostino Sarubbo 2023-03-07 08:09:21 UTC]

(In reply to Holger Hoffstätte from comment #6)
> I had it on profile-default (unset) and therefore enabled per-package. I
> explicitly disabled it globally, rebuilt llvm/clang/lld-toolchain-symlinks
> and tried again - no change, still no warnings from ar.


(In reply to Agostino Sarubbo from comment #8)
> ci has reproduced this issue with version 1.3.1-r1 - Updating summary.


Since I have reproduced with 1.3.1-r1 I doubt you have really disabled -native-symlinks

[Comment 10 Sam James 2023-03-07 08:15:11 UTC]

(In reply to Agostino Sarubbo from comment #9)
> (In reply to Holger Hoffstätte from comment #6)
> > I had it on profile-default (unset) and therefore enabled per-package. I
> > explicitly disabled it globally, rebuilt llvm/clang/lld-toolchain-symlinks
> > and tried again - no change, still no warnings from ar.
> 
> 
> (In reply to Agostino Sarubbo from comment #8)
> > ci has reproduced this issue with version 1.3.1-r1 - Updating summary.
> 
> 
> Since I have reproduced with 1.3.1-r1 I doubt you have really disabled
> -native-symlinks

Are you saying it works for you when you *enable* native-symlinks?

[Comment 11 Agostino Sarubbo 2023-03-07 08:19:49 UTC]

(In reply to Sam James from comment #10)
> Are you saying it works for you when you *enable* native-symlinks?

yes, this is a -native-symlinks issue. I don't know if the issue is what we are seeing in the summary or it is likely hidden.

[Comment 12 Agostino Sarubbo 2023-03-10 12:15:27 UTC]

ci has reproduced this issue with version 1.3.1-r2 - Updating summary.

[Comment 13 Agostino Sarubbo 2023-04-16 08:40:22 UTC]

ci has reproduced this issue with version 1.3.1-r3 - Updating summary.