@DimStar77 A little bit different as we have discussed, but maybe this is a better solution?

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 69.37%. Comparing base (08069f7) to head (1b8b504).
Report is 1 commits behind head on main.

@@           Coverage Diff           @@
##             main     #326   +/-   ##
=======================================
  Coverage   69.37%   69.37%           
=======================================
  Files          18       18           
  Lines         937      937           
  Branches      267      267           
=======================================
  Hits          650      650           
  Misses        174      174           
  Partials      113      113           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@DimStar77 A little bit different as we have discussed, but maybe this is a better solution?

Could probably work. Not sure though if just disarming stricter (not wrong) compiler behaviour for our own sake is the right approach

Ok, because the proposed empty map check just fails on my system

@DimStar77 A little bit different as we have discussed, but maybe this is a better solution?

Has this in fact been checked at all?

Because it very plainly CANNOT have any effect on the reported bug at all, which was about the line

if cc.has_multi_link_arguments(vscript)

producing the word "NO". Therefore the build is configured to not use symbol versions.

This PR still reports that LLD does not support version scripts (the map file fails to test-link) at configuration time and then passes an unused (no version file is being used) -Wl,--undefined-version argument at compilation time

$ LDFLAGS="-fuse-ld=lld" meson setup builddir/

[...]

Compiler for C supports link arguments -Wl,--undefined-version: YES 
Compiler for C supports link arguments -Wl,--version-script,/tmp/libproxy/src/libproxy/libproxy.map: NO 

As I said in the linked issue:

This is a configure probe check. It is checking to see if the linker (LLD) understands how to interpret a --version-script. That's entirely fine, as long as the version map is also passed when compiling the libproxy.so shared library.

Unfortunately LLD has decided to do the linker equivalent of compiling with -Werror=unused-function. Specifically, ld.bfd's -Wl,--no-undefined-version is the default for LLD. [...]

You could possibly go with just:

if cc.has_link_argument(vscript) or cc.has_multi_link_arguments([vscript, '-Wl,--undefined-version'])
  vflag += vscript
endif

Maybe not the most pretty to have double check but works.

Ok, because the proposed empty map check just fails on my system

Can you give some information about this failure? The empty.map file should look like this:

LIBPROXY_0.4.16 {
};

LIBPROXY_0.5.5 {
} LIBPROXY_0.4.16;

I do not have this linker on my system, so i proposed this PR and asked for feedback. I'm sorry that it does not fit your needs, but I'm sure we will come up with a good solution. Is the solutions from @jpalus fine then?

@TijlCoosemans I actually had a real empty file as we have just pushed in the last PR.

Is the solutions from @jpalus fine then?

Yes. And you do need both parts of the "or", due to binutils < 2.40: https://github.com/bminor/binutils-gdb/commit/27fb6a1a7fcd047e7d01bef8e3c12ed67bc7aed0

@TijlCoosemans I actually had a real empty file as we have just pushed in the last PR.

That gives the error ld: error: .../src/libproxy/test.map:1: unexpected EOF. It works if you add TEST {}; to test.map.

@trisk Could you please check whether the suggested change is also fine for Solaris?

@trisk Could you please check whether the suggested change is also fine for Solaris?

It works if I put TEST { local: *; }; in the mapfile.
Otherwise the default disposition of symbols is unspecified:

Undefined                       first referenced
 symbol                             in file
main                                /var/tmp/testfile-2c26bd.o  (symbol has no version assigned)

@trisk Could you please check whether the suggested change is also fine for Solaris?

It works if I put TEST { local: *; }; in the mapfile. Otherwise the default disposition of symbols is unspecified:

TEST { local: *; }; is okay here as well.

I've pushed a new version, could everybody please check their systems? In case this one is ready I'm going to release a new version

Looks good to me.

Works here too, but you can remove or cc.has_multi_link_arguments([vscript_test, '-Wl,--undefined-version']). There are no undefined versions or unknown symbols in test.map.

Works fine for me too with both lld and mold, although variant with '-Wl,--undefined-version' worked fine in both cases as well.