On Thu, 2020-12-17 at 17:11 +0100, Samuel Thibault wrote:
> Svante Signell, le jeu. 17 déc. 2020 15:54:28 +0100, a ecrit:
> > # define CLOCK_REALTIME_COARSE 5
> >
> > The problem with that option is that it is not yet supported,
> > resulting in EINVAL and a crash message on the console:
> > /hurd/crash: host <address> crashed ...
>
> We could comment this out indeed.
>
> But the program assuming that "it's defined so it works" is
> completely bogus on Linux as well: one may run an application built
> against a recent libc, on an old Linux kernel...
You are right, but how common is this combination? According to the
manpage of clock_gettime(): CLOCK_REALTIME_COARSE (since Linux 2.6.32;
Linux-specific) And that kernel was released on December 2, 2009.
> > Maybe even the stubs, like: posix_fadvise, acct, getrandom,
> > msg{ctl,get,rcv,snd}, sem{ctl,op,get}, vhangup,
> > pthread_getcpuclockid,
> > etc from gnu/stubs.h should be hidden too?
>
> ./configure knows that they are stubs, and thus correctly detects
> them as not being available.
Got it! Then we should concentrate on finding stuff defined in header
files, but not yet implemented in Hurd.
BTW: Any ideas about the second crash written about in my previous mail
that you edited away?
Thanks!
