Re: Open/pending issues as of Jan 17

[Rich Felker <dalias@libc.org>]

On Mon, Jan 18, 2016 at 12:25:30PM +0100, Szabolcs Nagy wrote:
> random stuff:
> - linux 4.* uapi (includes mlock2, but a lot of other stuff too)

Done.

> - symbol versioning (for at least gcc-6 libgcc_s.so)

This is a fairly invasive change and needs testing and possibly
further discussion. It's big enough I'd like to allocate it attention
on the roadmap for a specific release cycle. It's tentatively a
secondary target for the next cycle but we could move it up to primary
if this is becoming a serious problem.

> - __cxa_thread_atexit_impl ? (i expect this to come up in real code)

It looks to me like gcc emulates it with a pthread key to get a dtor
and doesn't actually need help from the libc. If so I don't think it
makes sense to bother with doing a better implementation on the libc
side unless/until we can resolve the lazy allocation bug in their ABI.

> - stdatomic.h

Let's look at this later.

> - string functions correctness

I don't think there's much that can be done aside from adding the
may_alias attribute. But at least that small improvement can go on the
agenda for the next release cycle.

> - regex bre extensions \+ \? \|

Possibly going to apply this now unless you see a risk of breaking
anything else.

> - regcomp increase tre_stack_new limit

Sounds ok to apply now but I lost track of where the patch is.

> - str{f,p}time %z %Z %k %l ?
> - noreturn __assert_fail ?
> - hidden visibility for internal functions

Let's look at these after release.

> - compiler builtins in libc (sqrt, fabs, memcpy..)
> - asm -> inline asm

A little bit of this was done (for arm) but I'd like to continue where
it makes sense.

> - ld128 math

Yes, that would be nice. Should we set a specific roadmap goal for
adding it or at least adding the most important stuff (like sqrtl)
that's currently badly broken.

Rich