Looks reasonable. I'll be waiting for kargl to say something, too.

Two changes will be made before landing this:

Nice submission! I'll find some time in the next days to work on this one. Please note that kargl thinks fmaximum and fminimum should never raise FE_INVALID, so some slight rework is needed (we do return (a+b) which triggers FE_INVALID if a or b are signalling nans). He has also asked that the build scripts be in alphabetic order. I have prepared patches for both which I will land in the next days, you can then rebase your feature based on that.

Once we get fmaximum_num and fminimum_num in, we should recommend that those be used instead. CC @jesuscblazquez_gmail.com.

Without having tested this patch, it sure looks like it would sidestep the problem elegantly.

@ngie Is it better now with comments?

This seems to have broken the build when building natively with WITHOUT_CROSS_COMPILER=yes, see thread on src-committers-all:

This looks pretty good.

Now fails to build like so:

In D56125#1283796, @haraldei wrote:

For some reason pkg thinks version 26+35.1_1 is greater than 26+35.2. Not sure how this will affect upgrades or the pkg builders?

In D56092#1283289, @haraldei wrote:

@fuz If you have an ARM box I can use, that would be great. Otherwise I can set up a VM too. I will either have to work my way up from openjdk8, or see if I can use a jdk for another arch to bootstrap it. The latter should be possible, but I haven't done that before, so will have to explore a bit.

Mostly AFK next week, though, so not sure how much I manage to do before then.

Fails to build on FreeBSD 15:

Would it be possible to generate the same bootstrap for arm as for the others? I can give you access to an ARM box if you need.

• libc/strrchr_test.c: explain what the unit tests test for

This test is an adaption of the memrchr test with minor modifications.

Do you plan to follow up on this before the start of GSoC?

LGTM

LGTM.
Please add a link to the changelog to the commit message when you land this one.

I would get rid of the type changes, as I don't think they actually improve the code. The end_result change is wrong.

Looks pretty reasonable, though I'll let Steve Kargl have the last word.
Needs a man page.