I updated the reset CR code.illumos.org/c/illumos-gate/+/3587/2 after a long delay. I'm not entirely happy with it.

→ CODE REVIEW 3587: 16643 sys/regset.h defines are unhelpful (NEW) | illumos.org/issues/16643

It seems like cleaning up ucontext.h and sys/regset.h has so many constraints that it's hard to come up with a solution that improves things and doesn't cause problems somewhere.

One approach I like would be to move to structs more like what the *BSDs do (no index name defines needed) , but that might break a lot of code that assumes a registers arrah.

A less ambitious approach would be to just try to mostly match the REG_* index names GLIBC presents, though we have a mix of 32-bit and 64-bit names (which have conflicts between the two sets) -- something GLIBC doesn't deal with.

What's currently up is mostly that (less ambitious) approach, with the 32-bit REG_ names presented as REG32_ to avoid conflicts. Not perfect, but seems to cause the least hassles so far.

you might find this interesting : <extrowerk> New article: blog.daveeddy.com/2026/07/06/illumos-doors

side note: in addition to validating parameters, if you're not planning on passing fds, you want to make sure you explicitly indicate that so your server doesn't end up with a bunch of unexpected fds sitting around

(ask me how I found out :P)

oh dear

thankfully it was during development :)

or i guess more i probably just forgot about the flag until i was like 'why is that happening?'

'oh'

This programme is full of file descriptors!

'my god, it's full of sta^Wfile descriptors'

what did you ^W out to add that

oh, it's a play on a quote from 2001 (the book/the movie sequel)

stars

ah

or was it also in the original movie? been a few decades since i've seen it :)

a friend of mine on the mastodon machine has in her bio a quote from a friend of hers calling her "full of golf balls"

the meaning of this I could not even begin to guess

jbk: only in the book of "2001"

and in movie 2010 (at about the beginning).

So I can not make RAM disk inside the zone ? I want to keep there perishable zfs encryption keys

and zfs dataset is delegated to a zone

[illumos-gate] 18221 mdb: 'htable.ht_pfn' may be used uninitialized -- Toomas Soome <tsoome⊙mc>

now it is just code.illumos.org/c/illumos-gate/+/1446 left - only cleanup of makefiles.

→ CODE REVIEW 1446: 13755 mdb: remove CNOWARN_UNINIT (NEW) | illumos.org/issues/13755

Anyone (maybe richlowe?) have suggestions for me to help me figure out how math.h changes could cause libm or libm1 symbols that ARE in the mapfile to not actually end up in the library? (something seems fundamentally broken if that's possible...)

what does output from gcc -E tell?

I figured it out. It's the inline definitions. I'm updating the issue.

yeah, that will make the global symbol disappear..

there is a way in the C spec to force an implementation to be emitted in a .o file for things that don't get the inline definition

when I did that, it was met with extreme suspicion, but maybe people will be more receptive now

IIRC, you stick `extern inline funcname(...);` in the .c file where you want the non-inline version to go

but it's been a few years

No symbols actually disappeared.

New headers expect something not exported by the ancient library.

[illumos-gate] 17689 posix_spawn() could avoid copying data and stopping threads -- Andy Fiddaman <illumos⊙fn>