On Monday 22 October 2012 04:06:24 Karel Zak wrote:
> On Mon, Oct 22, 2012 at 02:01:01AM -0400, Mike Frysinger wrote:
> > On Thursday 18 October 2012 16:06:28 Sami Kerola wrote:
> > > On Mon, Oct 15, 2012 at 3:14 AM, Mike Frysinger wrote:
> > > > On Sunday 14 October 2012 16:20:59 Sami Kerola wrote:
> > > >> POSIX.1-2001 declares usleep is obsolete.
> > > >
> > > > this is true, but it seems like a better answer would be to add a
> > > > usleep test to configure and provide a local fallback using nanosleep
> > > > if it doesn't exist.
> > >
> > > Is that necessary? There has been for example in old mount since March
> > > 2007[1] nanosleep() call, and I cannot remember anyone complaining it
> > > causing problems.
> > >
> > > [1] commit dc8fdc57cd3ba0658cf4ab05031695c2d2965f93
> >
> > i think you misinterpreted my objection. i don't have a problem with
> > calling nanosleep() -- when it makes sense. replacing a simple call to
> > a function that, while no longer part of the latest standard, was
> > mandated for many many years, and will most likely never be removed from
> > C libraries that have already been providing it (since it'd be an ABI
> > break), with a more complicated call for no real reason is pointless
> > imo. further, you'd be fighting a losing battle: developers will most
> > likely be working & testing on a glibc system where usleep does exist
> > and works fine, so they won't notice if it were added again.
> >
> > hence i suggested a trivial middle ground that is future proof and
> > doesn't penalize systems that do include usleep (i.e. glibc i.e. what
> > the majority of people run): if you actually have a system that lacks
> > usleep, then add usleep to the AC_CHECK_FUNCS tests in configure.ac, and
> > then add the simple replacement to include/c.h:
> > #ifndef HAVE_USLEEP
> > static inline int usleep(useconds_t usec)
> > {
> >
> > struct timespec waittime;
> > waittime.tv_sec = usec / 1000000L;
> > waittime.tv_nsec = (usec % 1000000L) * 1000;
> > return nanosleep(&waittime, NULL);
> >
> > }
> > #endif
> >
> > now everything should "just work".
>
> Exactly this solution we use for now (since 2009):
heh. too fast!
> $ cat include/usleep.h
>
> #ifndef UTIL_LINUX_USLEEP_H
> #define UTIL_LINUX_USLEEP_H
>
> #ifndef HAVE_USLEEP
> /*
> * This function is marked obsolete in POSIX.1-2001 and removed in
> * POSIX.1-2008. It is replaced with nanosleep().
> */
> # define usleep(x) \
> do { \
> struct timespec xsleep; \
> xsleep.tv_sec = x / 1000 / 1000; \
> xsleep.tv_nsec = (x - xsleep.tv_sec * 1000 * 1000) * 1000; \
> nanosleep(&xsleep, NULL); \
> } while (0)
> #endif
>
> #endif /* UTIL_LINUX_USLEEP_H */
if you were to do something like:
int foo = 10;
usleep(foo++);
this define macro would misbehave because it expands x twice. not that the
current code base does.
> ... so I don't think we have to check change anything.
indeed
-mike
Attachment:
signature.asc
Description: This is a digitally signed message part.
