Core Minutes 12/11/2012

• ScienceTools: (Jim) There were several items:

  • Bill wanted access to psf for comparing to event-by-event covariance, so Johann made irfs available by moving some general classes from latResponse to st_facilities adding st_facilities to GR and making use of it in AnalysisNtuple.
  • Work on flight FT2 for handling overlays:
    • Leon encountered some problems, now fixed
    • Will need to update astro::PointingHistory to handle candidate event times that lie outside of the intervals in the FT2 file, e.g., those during SAA passages. Easiest to have it throw an TimeErrorException and then have client code handle the exception.
  • IRF naming needs to be sorted out with C&A
  • Jim talked to Mike Corcoran about time-dependent irfs and HEASARC conventions caldb.

  See Science Tools Development Notes for details.

• FSSC: (Alex) is compiling the minor changes we made to ScienceTools source since v9r30p1. He'll be sending them to Jim so that they can be folded in at SLAC.

• Hardware and migration report (Tom G.)

  • nfs migrations are all complete (old servers still running until Jan 2013)
  • Batch machine increment has been ordered (not yet delivered)
  • The 10 new application servers, fermilnx, are up and running. All that's left to do is migrate services to them.

• Pass 7/8 (Leon) has been working on Overlay and friends and meanwhile uncovering bugs in SCons supersede handling.

  New merit variables have been added. The work is done; remains to be seen if there are unexpected repercussions.

  There were oddities in FT2 as viewed from inside Gleam.

  He is looking into a crash which only occurs after ~90,000 events — will not be easy to debug since it doesn't occur if you go straight to the problematic event without reading the others first.

  (Tracy) We're heading for a new tag, 20-08-00. Unlike the previous one, which was primarily made for infrastructure reasons, this one is content-full. It will be used to generate substantial MC datasets. The original plan was to freeze code by last Saturday, but at the P8 meeting Thursday Bill showed there were serious problems with energy assignment:

  • wrong total energy was being used
  • wrong division among tracks. Bill proposed just splitting 50-50 since we can't reliably do a better job, and even for 1-track events giving that one track just half the total energy (Tracy proposed assigning somewhat more in that case)

  Meanwhile, people discovered various other things needing fixing. This is now mostly done; hoping to finish up today and make a new HEAD. Heaather already made one with the bulk of the changes over the weekend and it basically looks ok. Systests are taking a long time, though. In particular event energy recon has gotten very slow.

• Core talk:

  • Systests (Heather) As of 20-07-00 (new ROOT version) certain systests have gotten very slow. She now runs those in the XX-long queue; the rest are in X-long so we can get as many results as possible sooner. For newer builds CalRecon will impact Systest times further.

  • ROOT and ST (Heather) has built the new ROOT version on all platforms now so it is available for ST. It's in use in LATEST and seems ok.

  • FSW release (Heather) Jana reports that the new build has been loaded. One run using it is planned for Dec. 18. Since no one has come up with reasons why we must be in sync and since another release is planned for early next year, we won't try to make an obf tag based on this release. We will for the next one, though.

  • SCons RM (Heather) Mac builds had been stuck. Restarting bldmac02 unstuck them to a degree. Since only one job can go at a time it will take a while to catch up.

    (Joanne)Large numbers of unit tests had been failing for Windows GR for a while — in fact, since support for non-installed headers on Windows with supersede had been added. The same programs run interactively would complete successfully. This was because RM does its own environment setup, at least on Windows. The contents of setup files generated by SCons had changed to support the new features but the code in RM hadn't kept up. Tom S. has now patched it; we expect this will fix the problem [and it did in the next GR LATEST].

    (Heather) The file protection problem for Windows zip files written to u52 is still with us. Our guess has been that there is a problem with the way the disk is mounted, but the Computer Center is now saying that glast-win04 may have a memory problem.

  • Developer environments (Joanne) Leon encountered a show-stopper problems using supersede on Windows, now patched. Certain relative paths occurring as part of CPPPATH for compiles were being misinterpreted (as relative to the wrong thing). The fix was to use absolute paths.

    Johann had a different problem with supersede on Linux, which came about because facilities was among the packages in his supersede directory (not something one would normally do since, to be safe, everything using it would also have to be in supersede, but he had only a few packages there). facilities has a special function, which is to create a header file containing all the packages of interest to the build, which is then used to define environment variables for those packages. The wrong (too small) set of packages will be used if facilities is in supersede. (Toby) That functionality should be in a separate package. (Joanne) This is what Navid originally suggested and, in retrospect, he was right.

previous

minutes index

next

J. Bogart, Last Modified: