Re: relaxing xnanosleep

[Jim Meyering]

Eric Blake wrote:
> Right now, xnanosleep is GPL.  But I don't see it calling exit() (which
> is normally the reason why most x*() interfaces are GPL), and it would
> be useful to use in libvirt.  Any objections to relaxing the license to
> LGPLv2+, including a technical reason I might be overlooking?  nanosleep
> and sigaction would also need relaxing from LGPLv3+ to LGPLv2+, but
> those are easier to justify.

Relaxing those is fine by me, but nanosleep has two problems:
- it is inappropriate for library use, since it manipulates signal masks.
- its implementation is incomplete:

    /* Suspend execution for at least *REQUESTED_DELAY seconds.  The
       *REMAINING_DELAY part isn't implemented yet.  */

    int
    rpl_nanosleep (const struct timespec *requested_delay,
                   struct timespec *remaining_delay)
    ...

      if (suspended)
        {
          /* Calculate time remaining.  */
          /* FIXME: the code in sleep doesn't use this, so there's no
             rush to implement it.  */

          errno = EINTR;
        }

But both of those are in order to handle being suspended a la control-Z,
and in a library context that might not be a concern.
Hence, it might be better to optionally stub out the suspend-handling
for applications that would prefer no signal handling interference.