fix(similarity): skip empty fingerprint sort - #1368

Thank you for this — and doubly so for reporting the issue and bringing the fix.

This is exactly the shape a fix should have: one file, one guard, and a comment that explains the why rather than restating the code. The distinction you drew in #1367 is the right one — zero elements suppress comparator calls, but that has never made a null base a valid array pointer, and count > 1 also skips the one-entry sort that could never have changed anything.

I checked the caller before merging: entry_count < MIN_FP_ENTRIES returns early via free(entries), and free(NULL) is well-defined, so the empty path stays correct end to end. Determinism at two or more entries is untouched.

Merging. Thanks again for the care you put into both the report and the patch.

One thing before I merge: this is still marked as a draft, so I have held off rather than promote it myself.

If that is just left over from opening it, mark it ready (or say the word here) and it goes in as-is — the review above is complete and CI is 28/28 green. If instead you are still polishing something, take the time you need; there is no rush from our side and I would rather merge the version you consider finished.

Thank you for the focused empty-fingerprint guard and for marking the PR ready. I have routed it at high priority in 0.9.1-rc because it removes undefined behavior from the indexing pipeline. It is back in the maintainer review queue, which is currently full, so the final re-review may take a little time.

Thank you for this one — a textbook fix: exact scope, a clear rationale comment right at the call site, and an honest write-up of why the empty path violates the qsort contract even though every libc happens to tolerate it. The determinism contract for count >= 2 is untouched, so this is behavior-identical where it matters and strictly more correct where it doesn't. Much appreciated, as always!

One note: we verified that no C-level assertion can bind this fix (the existing simhash suite already drives the empty path, but only a trap-on-UBSan build would ever fail without the guard). We're taking a follow-up on our side to bind it in the sanitizer lane so it can never silently regress.