Core Minutes 4 December 2007

Core Minutes 12/4/2007

Present: Joanne Bogart, Anders Borgland, Toby Burnett, Emmanuel Cephas, David Chamont, Eric Charles, Jim Chiang, Marek Domaracky (EVO Support), Richard Dubois, Warren Focke, Tom Glanzman, Navid Golpayegani, Heather Kelly, David Landriu, Chuck Patterson, Igor Pavlin, Leon Rochester, Tracy Usher, Eric Winter

Swig on Windows (Tracy) ran into a problem on Windows building packages like facilities and mootCore which invoke swig to provide a python interface: they complain about missing libraries. Jim Chiang found out this happens when the _debug flag is specified. This causes Windows to look for python debug libraries, not part of our distribution at present. This doesn't come up in RM builds since there the _debug flag is not set (nor, more mysteriously, on Glast TS). The _debug flag is useful when, for example, Tracy wants to be able to step into G4 code. Possible solutions are
1. Modify a header file in our swig installation in such a way that the debug libraries are no longer required when _debug is set
2. Add python debug libraries to those we distribute
3. (Jim) or could handle it with a post-process step, requiring changes only to SwigPolicy
(Heather) votes for #3.

(Toby) mentioned the related issue that CMT alone does not enforce proper build order for these swig-using packages. Navid has put some extra logic in RM to handle this, but MRvcmt, for example doesn't. facilities and mootCore are building correctly on Windows, but this is just luck.
SCons: (Navid) is still waiting to hear from package owners about 3 remaining ScienceTools packages (Jim reported that one of the 3, sane, has been tagged and a second, is nearly ready). Then he will be able to test SCons checkout and build from the real CVS repository.
GlastRelease/Big run: (Richard) We're off, using GlastRelease v13r5p1. We have gotten our double allocation (800 cores rather than 400) but mostly, for various reasons, we have not achieved that kind of usage . We've endured an Oracle glitch, hung-up xrootd servers, etc. Using the rhel4_gcc34opt build leads to an unacceptable percentage of crashes — signs (unconfirmed) seem to point to a problem in CalCalibSvc, which Zach has been attempting to reproduce in a controlled way — and also to surprisingly poor performance. While we try to understand these problems we've reverted to using the rh9_gcc32opt executable. By tomorrow may have v13r5p1 installed on Lyon, at which point we should be able to run backgrounds there.
Enough of the machinery of running; on to content:
- Bill has blessed ACD performance as adequate for this stage.
- Julie examined plots concerning SAA handling. The hole (while instrument is in SAA) is smaller and a somewhat different shape. This has now been fixed. The differences are not serious for background runs.
- Changes observed in magnetic latitude after recent code updates have been examined by Markus. He believes the changes are consistent with expectations.
- Before the Sky run, need to look at all-gammas to determine if Pass 5 is good enough. Toby is running all-gammas. He had to make patches to AcdUtil and CalibData to avoid crashes on Windows [Post-meeting (Joanne) confirms there was a 1-word overwrite error in CalibData but believes the AcdUtil change was unnecessary and probably harmful on RHEL4.] He's seeing unexpected behavior in effective area
- Recent ACD updates were described in detail by Eric C. He points out that there is no straightforward way to select the proper set of calibrations for a given run. One collection is needed for MC, another for real data. Certain kinds of consistency must be enforced. He would like to maintain job option details for this in ACD package territory but have access to a global variable to determine if the run is MC or not.
See the blog for the day-by-day gory details.
ssh: By now just about everyone has negotiated the change. Most have been able to use the Workbook instructions for Windows or Unix.
Skimmer(David C.) has installed a new version of the Skimmer which is compatible with ROOT 5.16.
Composite Event Lists (Heather) CEL is now in HEAD. (David C.) As far as he knows it's working; he's still testing.
Anders' concerns revisited (Heather) EFB blob and hadd issues mentioned last week are not yet resolved.
AOB
- Richard's hip is due to be replaced Friday AM. Expected hospital stay is 5 days. During the first three he will most likely be in a drug-induced haze, but after that, well, the hospital does have wireless and he's been warned it's likely to get boring. [But take it easy! ed.]
- (Toby) Bill, with help from Leon primarily and also Toby has resurrected the original gui to get at the replay feature. We should take care not to accidentally get rid of it in a fit of over-zealous clean-up. (Leon) uses the old gui on occasion for one reason or another.
- (Richard) Tracy will be spirited away to a boat in Florida Wednesday through Sunday. He might show up online in the evenings.
- (Richard) As we approach launch, greater formality in bug/task tracking is appropriate. Towards this end Tony has made a new JIRA project and suggests relevant issues be linked to its parent issue, MCT-1.

minutes index

J. Bogart, Last Modified: 11-Dec-2007 12:03:09 -0800