Core Minutes 1/31/2012
Core Minutes 1/31/2012ScienceTools: (Jim) added keywords to gtexpcube2 output. See Science Tools Development Notes for details. Also in progress is validation for unbinned integration over discrete diffuse source maps.
FSSC: (Tom S.) spent some time at Goddard last week training the new guy, Alex. Eric has been working on the new version of ST. It's coming along.
Pass7 reprocessing (Tom G.) There are two ongoing projects:
(Leon) Sasha produced a huge (50-60) batch of calibration files which need to be entered in the db. Entering them one by one via rdbGUI would be a tedious and error-prone process, so he and Joanne will team up to create a script containing the requisite MySQL commands. But not for a couple days. First, he, Johan and M-E will be bringing the skim of stacked AGN's up to date (will about double the size) and reprocess. They hope and expect to see the same kind of improvement Art saw with a smaller sample.
GR/Pass8 (Tracy, Leon) Concerning truncation: there is a list of hurdles to be overcome before the engineering run can happen, including
(Tracy) In other news, mostly geared towards upcoming Face-to-face, Alex has been looking at background rejection. Bill is going to do a grand overview of the Event Level analysis for Pass 7 (which he authored) as a way of reminding us of what we learned before. After that he is working on the covariant errors with the ultimate goal of using event by event errors with pointing instead of the IRF's. In the meantime he has learned a lot about how to help determine which events are providing the best PSF. This is another presentation he plans to make.
Core talk:
GR and end of rhel4 (Heather) ..is nearly upon us. It goes away at the end of February. Before then GR main and L1 branches need to build and run on rhel5, implying SCons build. The main branch is working except for event display (which is close; may just take more fiddling with job options) with CMT builds on rhel4 and vc71, also SCons builds on rhel4 and rhel5. All gaudi updates have been merged into L1; most likely minor fixes will still be needed for it to all build properly.
v19r4p1gr03 is the latest tag on the Pass 8 (old gaudi) branch. Liz has been asked to run systests, to be compared with a new main branch head tag. Other planned comparisons: CMT versus SCons builds on rhel4 for main branch, then compare rhel4 and rhel5 SCons builds. Then do the same for L1. Liz has most of the code in place to run systests semi-automatically for SCons. If it doesn't come together in time, we can do it by hand.
(Richard) M-E thinks she needs 2 weeks to prep L1 on her side.
(Richard) has asked John Bartlett what happens if we still need rhel4 after the deadline. Probably we'd get one box with strictly limited access. Perhaps we can buy support for a short time from Redhat.
Heather is keeping a list of everything involved in the final push.
Other GR news (Heather) The merit package is of no apparent use to anyone anymore and hence has been dropped from releases. (Joanne) That's for CMT builds. The same needs to be done for SCons builds.
u30 disk (Heather) It was up to 96% full, now down to 94%. This is where CMT builds are kept, including many pre-redhat 4 (rh9_gcc32 and rh9_gcc32opt for GR, ST, CHS and Beamtest). These are excellent candidates for archiving, along with tiger. Michael is working on this.
GR and vc90 (Joanne) There is a significant difference in build behavior between vc90 and vc71. There is a danger in mixing dll's linked against the debug and non-debug versions of the c runtime library in certain cases because they allocate memory from different heaps. To avoid any possible problems, with vc71 our standard procedure has been to always link against the non-debug crtl, even for debug builds. This is no longer possible with vc90. Code compiled debug — if it makes any use of the standard template library — must be linked against the debug crtl. And non-debug code must be linked against the non-debug crtl. This was causing mysterious problems with G4Generator or, rather, with the debug version of the G4 static libraries it depends on. The G4Generator library was requiring both the debug and non-debug crtl at link time, but the resulting dll fails to load. This may have to do with suspicious looking options in the link command generated by the standard (cygwin-based) procedure for building G4 on Windows. If that can't be easily changed, we could consider using Norm Graf's Visual Studio solution and project files. Unfortunately they are for VS 2010 and can't easily be downgraded to our VS version 2008. He will look into generating 2008 files for us after he returns from CERN next week.
Slow CMT RM builds (Heather) After some offline discussion, she has copied external libraries and perl scripts to the local disks of the 4 windows machines used for CMT RM builds. It should just take straightforward changes to the RM's database to make use of the local externals, so an easy thing to try, even if we don't expect a huge improvement. Getting more to be done locally would take more, and possibly more involved, database changes. Meanwhile, for whatever reason, recent builds have taken 10-11 hours, approaching reasonable.
SCons RM (Tom S.) It's generally working well. (Jim) triggered an optimized ST-HEAD build in RMII, but while the builds for the redhat platforms compiled, the unit tests did not run. (Tom S.) is investigating. [And later discovered the problem was due to a typo. He fixed the configuration entry and is retriggering the rhel5-32bit optimized build to make sure it works.] (Joanne) Windows LATEST builds have been failing on cvs checkout since Jan. 25th. (Tom) I'll have to look more closely at the Windows GR issue. It doesn't have an obvious cause.
New SCons version (Heather) You mentioned we could use an upgrade, right? (Joanne) So I did. I've forgotten if there was a specific reason, but it's probably a good idea in any case and in theory shouldn't do any harm. Will try out something newer on both Windows and Linux.
previous
|
minutes index
|
next
 |
J. Bogart, Last Modified: