[RFC] glibc AI policy. Round 3. Adopt the GCC AI policy.

Paul Eggert eggert@cs.ucla.edu
Wed Nov 4 06:17:29 GMT 2026


On 2026-11-03 14:49, Carlos O'Donell wrote:
> I think the leakage is in one direction only: outward.

Yeah, I would have thought so too a year or two ago, except that more recently I've seen leakage the other way, i.e., from tests to code - not in glibc, but elsewhere. It's not obvious leakage where you can see code copied from one spot to another; it's a leakage of such detailed info via the test interface that it goes beyond fair use even though it is not simple code copying.

> While glibc has more syscall wrappers, gcc has all of libstdc++ to test,
> libffi, the sanitizers (ubsan, asan, etc), libgomp, etc. all which
> exercise the host glibc and kernel in similar ways.

Yes, those parts of the GCC tests are like glibc tests, and this differs from the tests I was thinking about, which were the tests of the compiler proper and which are less like glibc tests. And I guess that means I am more concerned about the glibc-like part of the GCC tests, than I am about tests of the GCC compiler proper.

I hope this helps explain why I suggest applying the same rules to glibc's tests that we do for glibc's main code.


More information about the Libc-alpha mailing list