Monday Patch Queue Review update (2023-02-20)

Monday Patch Queue Review update (2023-02-20)

[Carlos O'Donell]

Most recent meeting status is always here:
https://sourceware.org/glibc/wiki/PatchworkReviewMeetings#Update

Meeting: 2023-02-20 @ 0900h EST5EDT

Video/Audio: https://bbb.linuxfoundation.org/room/adm-alk-1uu-7fu

IRC: #glibc on OFTC.

Review new patches and restart review at the top.

 * State NEW delegate NOBODY at 222
 * Carlos' SLI at 163 days.
 * Starting at 65235
 * Started discussing Arjun's failures on s390x.
  * Florian: Fedora builds against the z13 baseline.
  * Related commit: 22999b2f0fb62eed1af4095d062bd1272d6afeb1
  * Related commit: 7ea510127e2067efa07865158ac92c330c379950
  * Adhemerval: s390x build tests:
    * -march=z900, -march=z10, -march=z196, default
   * Review the earlier commits for impact on z13
   * Review build-many-glibcs configuration with -march= baselines for testing.
 * Adhemerval noted that Gentoo was discussing stable backport policy.
  * The open question is about hardware performance backports.
 * 65235: Added Redirects to longdouble error functions [BZ #29033]
  * Impacts ppc64le ieee754 long double. IBM to review?
 * build: All microbenchmarks should output in JSON. (bug 27740)
  * Ask Siddhesh to review.
 * Remap __GLIBC_FLT_EVAL_METHOD to 0 if __FLT_EVAL_METHOD__ is -1
  * Needs senior review for generic code. Will come back to this next week.
 * [v2] C2x scanf binary constant handling (Joseph)
  * Adhemerval to review.
 * implement dlmem() with audit extensions (Stas Sergeev)
  * Adhemerval: dlmem() needs some very specific requirements and memory cannot follow what dlopen will do regarding mmap flags.
  * Carlos: What is our use case here? Why?
  * Florian: Use memfd? Security review will eventually block fds since it can be used to inject created code.
  * Needs senior reviewer to discuss use cases.
  * Carlos will review the use case on list.
 * i386: Use pthread_barrier for synchronization on tst-bz21269 (Adhemerval)
  * Carlos to ask DJ to review for CI/CD impact.
 * Szabolcs: What about libmvec review?
  * Raising the minimum requirement to gcc 10. Distro gcc?
  * Florian: It is possible to build libmvec only if gcc is 10 or newer?
  * Historically libmvec has different versions and we had an Intel, AMD, and GNU ABI.
  * Some projects also implement what libmvec does in a single DSO to override.
  * Carlos: For rpm-based distros you could backport libmvec.so.
  * Florian: For libcrypt we define the function in the header and not always include it.
  * Adhemerval: Do we define the prototype?
  * Carlos: Yes, gcc has to know, and you add the pragma.
  * Szabolcs: No header magic yet. Explicit decls. Manually call the symbols.
  * Carlos: The risk is not at the distro level, because glibc 2.38 using distros will use newer gcc, so raising the baseline is less risk. The risk is in the glibc/gcc developer workflow impact.
  * Szabolcs: Making the libmvec.so non-optional seems safer.

-- 
Cheers,
Carlos.