[Developers] Announce API spec
schnetter at cct.lsu.edu
Tue Sep 12 15:30:12 CDT 2006
On Sep 8, 2006, at 12:41:28, Thomas Radke wrote:
> as I had announced earlier on this mailing list, I've been working
> the last couple weeks on a specification for a generic Cactus API to
> announce simulation metadata. Attached is my first draft of it.
> Like many (most?) other software developers, I had to actually write
> some code first in order to come up with a spec that is practically
> useful. So a prototype implementation exists, however I haven't put
> CVS yet. Before that can happen I'd like to post the spec here for
> discussion and get answers specifically to the following open
> 1) Is it okay to call my implementation of the Announce API, as
> proposed in this spec, thorn 'Announce' ?
> There exists already a thorn of same name in the Portal
> It provides functionality to publish metadata, but only to one
> specific type of external information service: a Cactus portal.
> Portal/Announce doesn't provide a generic API for application
> to announce arbitrary user-defined runtime simulation metadata.
I would use a different name, e.g. "MetaData". Also, having two
thorns with the same name is asking for trouble.
> 2) Do the Announce user API functions need a 'severity' level
> This was suggested in earlier discussions by Tom and Erik, in
> analogy to the warning level argument passed to CCTK_VWarn().
> I'm not sure how levels would be practically used in an Announce
> context. What individual levels should be predefined ? Should the
> level argument be interpreted as a bitmask rather than an
> integer ?
I would not add such an argument. One cannot assign severities to
events. Rather, there would be parameters for each announcing thorn
that then decide locally whether CCTK_Announce is called. One user
may be interested in horizons, the other in wave forms, the third in
I/O performance data, and it won't be possible to handle that via
severities. Bit masks would make it only more complicated.
> 3) Should the Announce user API functions take an additional argument
> 'CCTK_POINTER_TO_CONST cctkGH' ? It's not clear to me why announce
> callbacks would need access to a cGH structure.
I would add it just for completeness.
Erik Schnetter <schnetter at cct.lsu.edu>
My email is as private as my paper mail. I therefore support encrypting
and signing email messages. Get my PGP key from www.keyserver.net.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 186 bytes
Desc: This is a digitally signed message part
Url : http://www.cactuscode.org/pipermail/developers/attachments/20060912/2e4b6e67/attachment.bin
More information about the Developers