[DRMAA-WG] Torque/PBS DRMAA - drmaa_wi* POSIX equivalents I

[Peter Troeger]

> Issue:
>
> Using drmaa_wifaborted(), drmaa_wifexited(), drmaa_wifsignaled(), etc.
> is tedious and error-prone. For example, it's not clear what the right
> order of all the calls should be and what if none of them gives a
> meaningful answer, i.e. the job exit status could not be determined
> somehow. Some one-call semantics to get job exit status should be
> considered. Of course, this would kill all these formal POSIX  
> semantics
> of wait and wif* functions.
>
> -----------------------
>
> I agree here that the usage is not straightforward sometimes.  We  
> wanted
> to follow POSIX for conformity, so many developers are expected to be
> already familiar with the macros semantics.

This already triggered long discussions in the DRMAA group. The IDL  
spec looks like this:

--- snip

interface JobInfo {
  readonly attribute string jobId;
  readonly attribute Dictionary resourceUsage;
  readonly attribute boolean hasExited;
  readonly attribute long exitStatus;
  readonly attribute boolean hasSignaled;
  readonly attribute string terminatingSignal;
  readonly attribute boolean hasCoreDump;
  readonly attribute boolean wasAborted;
};
interface Session{
...
JobInfo wait( in string jobName,  in long long timeout)
...
}

--- snip

A language mapping can represent the output informations with several  
functions, like today in C DRMAA, or use the result structure as  
result of the call, like today in Java DRMAA. Therefore, the overall  
language-independent issue is solved, but for the DRMAA C binding it  
would demand a new version of the C API specification. Therefore, I  
suggest to open a new tracker item for the C binding.

Regards,
Peter.


>
> This is a complex issue that requires substantial effort, so we  
> need to
> mark this issue as the future versions work.
>
>
> -Hrabri
> --
>   drmaa-wg mailing list
>   drmaa-wg at ogf.org
>   http://www.ogf.org/mailman/listinfo/drmaa-wg