Am 20.08.2012 um 12:17 schrieb Peter Tröger <peter@troeger.eu>:
Dear all,
since the public comment period for DRMAAv2 C-binding is over, we need to discuss our reaction on the comments. The next DRMAA conf call is therefore scheduled for Wednesday, August 22, 19:00 UTC - 20:00 UTC.
1. Meeting secretary for this meeting?
Participants: Andre, Daniel, Mariusz, Roger, Stephan Klauck, Peter Meeting secretary: Peter
2. Status report by the participants
All in good health, some reporting from Peter about OGF35 in Delft
3. Public comments on DRMAAv2 C-binding (http://ogf.org/gf/docs/comment.php?id=388) - pointer NULLing by the library ? What it means for the interface ?
- No decision at OGF35, voting needed - Roger, Peter: "principle of least astonishment, no NULLing in the library" - Andre, Daniel, Mariusz: "security problem, asynchronous free problem, OpenMPI example" - 3:2 voting result for changing the API -> free* functions get pointer variable reference and assign NULL to it - Voting also clarified that no other API changes are made
- Problems with OStype and CPUtype enumerations
- No decision at OGF35, voting needed - 5:0 voting result for keeping MACOS identifier - OS version is already a separate concept in the API - 5:0 voting result for changing TRUE64 to TRU64 - 5:0 voting result for adding 64bit variations of all instruction set architectures - These changes demand an errata for the DRMAAv2 root specification
- Semantics of DRMAA2_DATA_SEG_SIZE
- No decision at OGF35, voting needed - 5:0 voting result for the proposal by Rayson Ho: DATA_SEG_SIZE is renamed to DATA_SIZE - Explanation must extended - this value refers to 'initialized data, uninitialized data and heap' - These changes demand an errata for the DRMAAv2 root specification
- Additional free functions needed
- No decision at OGF35, voting needed - 5:0 voting result for adding them
- Return type for wait functions
- OGF35 decision acknowledged by Roger - IDL spec should have void return type for wait() functions - Additionally, language bindings should be allowed to change the return type for chaining support - These changes demand an errata for the DRMAAv2 root specification
- _j_ vs. _job_
- 4:1 voting result: Keep it as it is now
- Small issues
- jobCategories should be accessible outside the job session - Daniel: Need scheduler connection for fetching the information, and then again in the reservation session - Mariusz: Making a reservation without some job submission afterwards is very unlikely - 5:0 voting for keeping it as it is now - Native drmaa2_string type, instead of char* pointers - OGF35 decision acknowledged by Roger - UNSET for JobInfo: 5:0 voting for adding it - Remove endIndex upper limit specification demand from the root spec - Demands an errata for the DRMAAv2 root specification I will now start to prepare both the errata document and the final C binding for the editors. Best regards, Peter.