On Tue, Jul 09, 2024 at 09:47:22PM +0200, Alejandro Colomar wrote:
> On Tue, Jul 09, 2024 at 09:36:34PM GMT, наб wrote:
> > On Tue, Jul 09, 2024 at 09:20:28PM +0200, Alejandro Colomar wrote:
> > > Anyway, for 2008, was it non-standard behavior?
> > It wasn't "non-standard" so much as it was left up to the implementation
> > (still is), except for timer_create().
> >
> > All implementations agree here, so there's no point in splitting hair IMO
> Okay. Then, maybe, I'd remove the "Linux and NetBSD" bits from the
> commit message, and just say something like all implementations do that?
Okay, wider implementation survey:
alarm setitimer timer_create
Linux SI_TIMER SI_TIMER SI_TIMER
NetBSD SI_TIMER SI_TIMER SI_TIMER
FreeBSD not not SI_TIMER
illumos not not SI_TIMER
OpenBSD not not N/A
(Well, OpenBSD doesn't have POSIX timers, so the value of comparing it
is questionable.)
I wouldn't call it a "Linux extension" because implementations are free
to pick anything for setitimer and alarm signals (just like they are free
to implement alarm in terms of setitimer).
But also clearly not all implementations do this.
But also /some/ non-Linux ones do.
But also saying something happens "Under Linux" here is goofy,
because these are linux man-pages.
Maybe NOTES: "POSIX only guarantees SI_TIMER for signals created by
timer_create(2). Implementations are free to also provide it for other
types of timers. The Linux behaviour matches NetBSD."?
Best,
Attachment:
signature.asc
Description: PGP signature
