I'm trying to wrap up a way-overdue release, and there are some known things I wanted to get done in this cycle that just need to be put off: - mntent fix: At least as I remember the state of things, there's no fix patch that's been written and tested that's not making additional possibly- (likely-) unwanted behavioral changes. - pthread_barrier_wait: It seems something is seriously wrong and I (or someone else, but this is probably best for me to do) need to sit down and work out the sequencing logic and figure out what's wrong. The result might entail a redesign and rewrite, so I don't think we can hold things up with this. I should probably drop use of it in musl (all uses can be replaced by simpler primitives) but I'd rather do that after release too. - ld-musl....path file location logic: There's a patch for this I haven't reviewed. It's a low-impact issue so I'll look after release. - ASM source file conversions to inline asm: I'm not sure of the status on this; it was in progress but not ready for merge before. - Some mess in x32 sys/user.h: A bunch of members are 32-bit and some (maybe all) need to be 64-bit? - Some RPATH $ORIGIN failure cases: I need to review this again. - Request for preadv2/pwritev2: No follow-up getting info on whether or how glibc exposes them. - CMSG_* warning mess: We may want to finally act on this since it keeps tripping people up, but I don't want to do a hasty change and risk introducing a bug at release time. Let's look at this later. - NFTW_CHDIR support: This is probably ok but I haven't reviewed it. Hope to merge after release. - nscd querying when the daemon is disabled for some query types: IIRC this had really problematic behavior we don't understand well still on the nscd side that might mean we're doing things wrong or that there's no way to do things right. Needs investigation before any action. - Glob altdirfunc stuff and related changes: I don't remember if the UB was all resolved; if so this may be acceptable to merge. - _SC_NPROCESSORS_CONF: Need to review where this is at. - Improvements to qsort: These were put off pending qsort_r, which is done now, so it may be appropriate to fix up and merge these. And some things that were open but approved for merge or trivial, that I'm fixing now: - Zeroing canary byte to avoid leaking it via string functions - Avoid runtime conversions of floating-point constants - __WORDSIZE inconsistency on x32 - Linux uapi additions - nice EACCES/EPERM issue All in all, the above list turned out to be a lot more extensive than I expected, and this really highlights the need for an issue tracker... Please reply to add anything I may have missed. If you have follow-up on any of the actual items, though, maybe hold off for a few days until I can actually get a release out and have mental space to deal with new stuff. :-) Rich
On Tue Mar 8, 2022 at 11:54 PM CET, Rich Felker wrote:
> Please reply to add anything I may have missed. If you have follow-up
> on any of the actual items, though, maybe hold off for a few days
> until I can actually get a release out and have mental space to deal
> with new stuff. :-)
the libintl patch i submitted (`abort transaction lookup if NULL is
passed as msgid1`) should perhaps be on the list too.
--
alice
On Tue, 8 Mar 2022 17:54:15 -0500, Rich Felker <dalias@libc.org> wrote:
> Please reply to add anything I may have missed. If you have follow-up
> on any of the actual items, though, maybe hold off for a few days
> until I can actually get a release out and have mental space to deal
> with new stuff. :-)
posix_spawn_file_actions_t packing is still pending, though I don't know
if it belongs on this list.
[-- Attachment #1: Type: text/plain, Size: 3320 bytes --] On Tue, Mar 8, 2022 at 2:54 PM Rich Felker <dalias@libc.org> wrote: > I'm trying to wrap up a way-overdue release, and there are some known > things I wanted to get done in this cycle that just need to be put > off: > > - mntent fix: At least as I remember the state of things, there's no > fix patch that's been written and tested that's not making > additional possibly- (likely-) unwanted behavioral changes. > > - pthread_barrier_wait: It seems something is seriously wrong and I > (or someone else, but this is probably best for me to do) need to > sit down and work out the sequencing logic and figure out what's > wrong. The result might entail a redesign and rewrite, so I don't > think we can hold things up with this. I should probably drop use of > it in musl (all uses can be replaced by simpler primitives) but I'd > rather do that after release too. > > - ld-musl....path file location logic: There's a patch for this I > haven't reviewed. It's a low-impact issue so I'll look after > release. > > - ASM source file conversions to inline asm: I'm not sure of the > status on this; it was in progress but not ready for merge before. > > - Some mess in x32 sys/user.h: A bunch of members are 32-bit and some > (maybe all) need to be 64-bit? > > - Some RPATH $ORIGIN failure cases: I need to review this again. > > - Request for preadv2/pwritev2: No follow-up getting info on whether > or how glibc exposes them. > from https://man7.org/linux/man-pages/man2/writev.2.html: preadv2() and pwritev2() first appeared in Linux 4.6. Library support was added in glibc 2.26. (we recently added them to bionic too.) - CMSG_* warning mess: We may want to finally act on this since it > keeps tripping people up, but I don't want to do a hasty change and > risk introducing a bug at release time. Let's look at this later. > > - NFTW_CHDIR support: This is probably ok but I haven't reviewed it. > Hope to merge after release. > > - nscd querying when the daemon is disabled for some query types: IIRC > this had really problematic behavior we don't understand well still > on the nscd side that might mean we're doing things wrong or that > there's no way to do things right. Needs investigation before any > action. > > - Glob altdirfunc stuff and related changes: I don't remember if the > UB was all resolved; if so this may be acceptable to merge. > > - _SC_NPROCESSORS_CONF: Need to review where this is at. > > - Improvements to qsort: These were put off pending qsort_r, which is > done now, so it may be appropriate to fix up and merge these. > > And some things that were open but approved for merge or trivial, that > I'm fixing now: > > - Zeroing canary byte to avoid leaking it via string functions > - Avoid runtime conversions of floating-point constants > - __WORDSIZE inconsistency on x32 > - Linux uapi additions > - nice EACCES/EPERM issue > > All in all, the above list turned out to be a lot more extensive than > I expected, and this really highlights the need for an issue > tracker... > > Please reply to add anything I may have missed. If you have follow-up > on any of the actual items, though, maybe hold off for a few days > until I can actually get a release out and have mental space to deal > with new stuff. :-) > > Rich > [-- Attachment #2: Type: text/html, Size: 4252 bytes --]