-----Original Message----- From: owner-drmaa-wg@ggf.org [mailto:owner-drmaa-wg@ggf.org] On Behalf Of Peter Troeger Sent: Wednesday, October 12, 2005 4:38 PM To: DRMAA Working Group Subject: Re: [drmaa-wg] Additional error for C binding

I am not happy with such an approach. The DRMAA spec cannot assume how the DRMS is building its job id. For example, we might have a DRMAA library for GT4 where job id's are really long url's. In this case it might happen that truncation is bad. The according library could create an own internal mapping (short id - real id), but as we all know this demands unwanted persistency in the library.

Giving a too small buffer for an information is a problem, and this should be made explicit. I vote for an error code.

Peter

Am 12.10.2005 um 23:28 schrieb Daniel Templeton:

...

I knew you'd think that. :)

I actually think that truncating the id is a better solution than returning an error code, *IF AND ONLY IF* that behavior is clearly described in the C binding specification.

Daniel

Roger Brobst wrote On 10/12/05 13:42,:

...

I think the spec should specify some constant like DRMAA_JOBID_BUFFER and the implementation should truncate the output string if a sufficient buffer is not provided.

I do not think this omission merits breaking the existing C-Lang API or rehashing the underlying discussion about memory management.

-Roger

In a previous e-mail, Peter Troeger wrote:

...

Oh, Dan is now back on US time. I am the only remaining person working at night on DRMAA ;-)

If we can agree on the last proposal from Dan, I could add an according test to the test suite. The first proposal would break all existing implementations and applications.

Hrabri, please but this on the agenda for the next phone conference, together with the other new issues from the last mails. Somebody might also add a tracker item.

Thank you, Peter.

Am 12.10.2005 um 21:44 schrieb Daniel Templeton:

...

I think the SGE implementation just truncates the job id, but I'm not sure. My vote would be, as Hrabri fell just short of suggesting, to drop the length parameter and make the buffer a char**. I don't think we have an error code which adequately represents the described error condition, but I'd say Invalid Argument is the best we can do with what we have.

Daniel

Rajic, Hrabri wrote On 10/12/05 12:20,:

...

There is how the spec treats a similar parameter job_name:

DRMAA_JOBNAME_BUFFER = 1024.

At the other place spec says:

job name A job name SHALL comprise alphanumeric and _ characters. The drmaa-implementation SHALL NOT provide the client with a job name longer than DRMAA_JOBNAME_BUFFER -1 (1023) characters. The drmaa-implementation MAY truncate any client-provided job name to an implementation-defined length that is at least 31 characters. The attribute name is drmaa_job_name

I think that the intent was to let the implementation provide adequate space to job_id since it was not marked as an In parameter. The app would copy it in an adequate length buffer (incurring performance hit). C-spec has done optimization here, but the adequate error code is now missing in the language independent spec.

Hrabri

> -----Original Message----- > From: owner-drmaa-wg@ggf.org [mailto:owner-drmaa-wg@ggf.org] On > Behalf > >

Of

> Peter Troeger > Sent: Wednesday, October 12, 2005 1:53 PM > To: DRMAA Working Group > Subject: [drmaa-wg] Additional error for C binding > > Hi, > > the DRMAA C-binding 1.0 defines an additional parameter for the > string length (job_id_out_len) whenever job_id_out is part of

...

...

...

...

...

> method signature. In the DRMAA spec, this is not considered -

1023 characters for job_id is quite generous insuring that the most majority of cases will work w/o problems. I would not be happy if the application has a bunch of successful runs and all of the sudden is refused job submissions because of the inadequate buffer lengths. If we advise users in experience documents to bump the buffer sizes for GT4 or similarly verbose DRMSs we could get away with error codes. Char** avoids truncations and errors, but precludes simple buffer arrays in favor of char ** arrays and indirect accesses. Using 1024 characters job_id arrays is also a poor use of memory, so we might be OK here to complicate slightly the recommended programming model for the users. We need to discuss this a bit more. Hrabri the the

...

...

...

...

...

> string is simply marked as output argument. I would be interested > what kind of error a DRMAA C-library should return of the > buffer for > the job_id_out parameter is too small: OUT_OF_MEMORY ? > INVALID_ARGUMENT ? > > Regards, > Peter. > >