It's not the uuid library, it's specifically uuid_generate_time_safe() (which exists on Linux, but not on OS X, even though OS X does have libuuid).

That looks autogenerated. The configure check works here on Linux.

With the patch:

checking for uuid_generate_time_safe in -luuid... yes

Now use a non-existing function name:

AC_CHECK_LIB(uuid, uuid_generate_time_unsafe)

checking for uuid_generate_time_unsafe in -luuid... no

I did not look at the autogenerated configure, but it seems to work here.

Is this comment copy-pasted from somewhere else? It doesn't seem to apply here...

The actual signature here is int uuid_generate_time_safe(uuid_t out).

This call should match the function's signature.

setup.py has a bit of code to find libuuid in the proper place (self.compiler.find_library_file(lib_dirs, 'uuid')). It sounds like you should replicate that in the configure script, but it looks like you don't?

I'm not sure about this. -luuid is only needed for the _uuid extension module, and it's already added by setup.py.

I don't think that it's correct to test the define value. Only check if it's defined:

#ifdef HAVE_UUID_GENERATE_TIME_SAFE

Yeah, I was about to write about that. I couldn't make autoconf set #define HAVE_UUID_GENERATE_TIME_SAFE 0 if uuid_generate_time_safe() isn't available yet. Working on it :)

What was broken about the original patch except that some autogenerated comments were suboptimal?

The existing AC_CHECK_LIB(dld, shl_load) should have all the same problems.

Is it possible to put directly the AC_DEFINE() here, instead of using a variable?

[ax_cv_uuid_generate_time_safe=no] would be replaced with [] in that case.

I tried that but I got the following warning:

configure.ac:2688: warning: AC_CACHE_VAL(ax_cv_uuid_generate_time_safe, ...): suspicious presence of an AC_DEFINE in the second argument, where no actions should be taken

Oh, I hate autotools, mysterious tools... Ok, leave it as it is ;-)