Re: [gpsd-dev] regression tests blow up on os x 10.9 with shm issues

[Greg Troxel]

Greg Troxel <address@hidden> writes:

> NetBSD man page says:
>
>      IPC_RMID   Remove the shared memory segment specified by shmid and
>                 destroy the data associated with it.  Only the super-user or a
>                 process with an effective uid equal to the shm_perm.cuid or
>                 shm_perm.uid values in the data structure associated with the
>                 segment can do this.
>
> which matches
>   http://pubs.opengroup.org/onlinepubs/9699919799/functions/shmctl.html
>

> but my latest run did not leave one lying around.  So maybe this was
> just one odd test that got fixed.

update: running tests *once each* at
  release-3.11-463-gb355c9f
with no shm segements beforehand, I find:

netbsd-5 amd64: passes, one leftover segment
netbsd-6 i386: passes, zero leftover segments
osx-10.9 amd64: passes, one leftover segment

Also ipcrm promptly removes the segments.  So I don't think it's
anything about refcounting and gc.  <wildguess>Perhaps all refs have to
be dropped, and if this is about the dropping thread/process being too
fast.</>

pgpLBD4jo79Uk.pgp
Description: PGP signature