[r3wretrhy]

Fixes #4019.

TraceStream.cc and launch_debugger.cc call basename() on a const char*, which relies on the GNU declaration that glibc puts in <string.h>. Outside glibc that declaration isn't there, which is the TraceStream.cc failure from the #4009 build log.

As agreed in the issue, both call sites now use the POSIX version: include <libgen.h> and hand basename() a copy of the path, since POSIX allows it to modify its argument. In make_trace_dir() the name is now worked out once, before the retry loop.

For ordinary paths the result is the same. I compared old and new on glibc for absolute, relative and bare names and got identical output. They only differ for a path ending in /, which an executable path or the temp script name can't be.

Testing: built rr on Debian 13 (GCC 14, glibc 2.41). rr record created true-0, true-1 and bn_test-0 under _RR_TRACE_DIR, so the trace directory naming and the nonce retry still work. This machine has no hardware perf counters, so recording stops right after that and I couldn't run make check.

@GitMensch you'd offered to do this one. It looked idle, so I put up the small version, but I'm happy to close this if you already have a branch.

[GitMensch]

LGTM - thanks for the PR, seems like I haven't checked that in at all...

I've had a quick moment of "why not setting the fixed part once - but the output of trace_save_dir() is cached inside and that's not a tight loop in the first place - still good that you've moved the copy + basename call out of it.

[rocallahan]

Was this produced by AI? If so, please say so according to the AI usage policy:

When submitting issues or PRs, please clearly indicate how LLMs were used. Issues or PRs which appear LLM-generated but carry no such indications may be closed without comment.
When submitting PRs for review, the submitter must be human and take full responsibility for understanding the changes, regardless of how LLMs were used.

[r3wretrhy]

Yes. An AI agent wrote this change, ran the build and checks described above, and wrote the PR description. I missed the policy when it was opened, sorry about that. I've now read the diff and I'm the person responsible for it; if you'd rather not take it on that basis, feel free to close it.

[rocallahan]

I'm not sure it's safe to modify the characters at &module_path[0] ... POSIX basename() might insert a null byte, changing the length of the string, which might confuse std::string. The GNU version is better because it never modifies the string. So we don't want this change.