ptribble - re 17453 (fenix) I think that's clarifying rather than muddying. I can stop looking at differences between the forks at least. The values from the test program vary on OmniOS/Helios but they seem to always be page aligned, unlike the other systems.

illumos 17453 fenix

BUG 17453: ed(1) segfaults on writing empty file (In Progress)

↳ illumos.org/issues/17453 | code.illumos.org/c/illumos-gate/+/4282

It's definitely a weird one! Glad I can help shine a little light on mapping this particular rabbit hole.

It's apparently due to omnios/helios having ASLR enabled on sshd. If I log in on the console, I see the same behaviour as tribblix et al.

.oO oops.

Could you try `pfexec psecflags -s aslr $$` in your shell and see if it changes things?

Segmentation Fault (core dumped) (on OI)

Bingo!

With that, reproduces on Tribblix.

I would never in a million years have expected this. what a journey andyf

kwaltz - thanks for reporting it. It's a real bug in ed(1) and my proposed fix is at code.illumos.org/c/illumos-gate/+/4282. I can give you a hotfix for OmniOS if you'd like, otherwise it will be updated next time OmniOS has a weekly release.

→ CODE REVIEW 4282: 17453 ed(1) segfaults on writing empty file 17455 Add CTF data to ed(1) (NEW) | illumos.org/issues/17453

thank you so much andyf! I'll be sure to grab the release when it comes out

Thanks for finding this @kwaltz .

heh.. i wonder how long that bug has been there

Oh, ed crashes the same way on Solaris 11.4, once I worked out the magic incantation (using sxadm) to disable aslr

oh, so it does

With C99 `bool`, is there a reason we couldn't get rid of `boolean_t` in illumos?

It's probably a bit tricky. It's part of the MAC interface for example.

Dammit, was mid-typing "existing interfaces?" and yeah, I guess so.

The ena driver has a nice mix of boolean_t and bool for that reason.. Well, I converted everything I could to bool.

"sizeof (bool) != sizeof (boolean_t)" which matters for existing interfaces (i.e. what I was typing).

I SAW that in ena(4D).

(I don't think we have any frequently-allocated structures where adding more <<HOLE>> is horrid, but that we have <<HOLE>> in things bugs me, probably irrationally, and it's now in my same headspace as precision-data-layout in a language...)

The boolean_t is part of the ABI for many things and where it's part of that ABI, it can't change.

Also depending on what you care about, you cannot build C89 software with bool present generally speaking, unlike other extensions.

rmustacc: we should probably have a sys/bool.h rather than stdbool, for uts, too?

or does sys/stdbool.h end up behaving fully?

headers that smell of userland scare me

[illumos-gate] 17307 ufs/newfs: dprintf() cleanup -- Hans Rosenfeld <rosenfeld⊙gho>

richlowe: <sys/stdbool.h> was the one that the kernel can safely use.

ah, just the 'std' giving me the fear

I think of it like <sys/stdint.h>.

[illumos-gate] 17444 gitignore doesn't know findunref -- Patrick Mooney <pmooney⊙pc>