Core Minutes 6/16/2009

• ScienceTools: See this week's report.

• FSSC: (Eric W.) Freeze went into effect yesterday as scheduled, with release v9r15p2. FSSC is poised to incorporate essential updates, if any.

• Documentation: (Chuck) sends the following:

  Since last week, Jim's updates to gtmktime, which added instruction for grouping components of filter string, and to the Unbinned Likelihood Tutorial, which incorporated the SmoothBrokenPowerLaw, have been posted to the web. He also has an update to gtfindsrc which will be posted later today. And, since we've accumulated quite a few updates during the past 6 weeks, it's time to create a new tag so Berrie can update the mirror site.

  Beyond that, I've been cluster-bombed during this past week with several gnarly problems such as one reported by Su Dong involving Kerberos, WinSCP, and PuTTy. I'm also revisiting some rather basic documentation problems concerning setting up the environment for running on SLAC Public.

• Pass7 reprocessing: (Tom) Main activity this past week, now complete, was to recover files which didn't make it the first time. We now have 4122 merit files and 4121 FT1. The missing file corresponds to a bad run.

  In the course of reprocessing several problems were discovered which are now fixed. There are still outstanding problems with ROOT, skimmer and ScienceTools.

  ROOT (Heather) A ROOT bug concerning cloning trees has been addressed in ROOT 5.20-gl4.

  ScienceTools (Tom) The ST issue concerns the package tip which is responsible for ROOT i/o. Under certain circumstances it generates a defective FT1 file without any notification of failure. (Heather) tip, like Gleam, turns off much or all of ROOT's native signal handling. There was some speculation that it would be well to re-instate it. However, she found a place in the tip code which could be responsible for the problem we're seeing: a return code from a particular operation is being ignored. (Tom) The problem is transient and rare (failure rate of about 0.1%) so not readily reproducible. He suggests we fix the known bug in tip first, rather than tinkering with ROOT signal handling. (Heather) will talk to James Peachey about a fix.

  skimmer (Heather)

  • The web front-end has been updated to use a new version which knows how to handle newer merit files with a header; however it cannot handle the old ones! A work-around is now in place so that the front-end can handle both.
  • Tracy ran into a limit on number of files to be chained. We could maybe get around this by using Composite Event Lists.
  • There may be a separate issue concerning duplicate event ids.
  • There will be a new skimmer soon, built against the new ROOT.

• Disk clean up: (Heather) Where are we w.r.t. 40 Gbytes of overlay files? (Leon) is waiting to hear they've been archived, then can get rid of most of them. He'll send a list of files to Navid.

• Addio RHEL3 (Heather) We have to rid ourselves of RHEL3 soon because

  • SLAC RHEL3 norics are going away at the end of the summer.
  • Newer versions of external libraries such as ROOT will not run on RHEL3.

  She proposes we eliminate any dependence on RHEL3 a month or so before we absolutely have to to allow time for gotchas. Among other things, this probably entails some work on Sys tests (seconded by Elizabeth) and Pipeline scripts. (Warren) What about the 160 RHEL3 farm machines? (Heather) We need to ask SCCS about that; perhaps Richard can inquire.

• Announcements: (Heather)

  • Richard is back, but swallowed up by meetings.
  • Emmanuel will be leaving on July 10.

• Core talk:

  • SCons RM and Mac (Navid) The afs token issue has been resolved somewhat. v9r15p0 built nearly successfully: only failures were in test programs which need some code changes before they can run properly under SCons. However we can only submit one job at a time. If more jobs are running simultaneously, the first to finish destroys the afs token, causing the others to fail.

  • SCons RM and RHEL5 (Navid) RHEL5 has gcc 4.0, which does not include libg2c (needed to supply run-time support to c modules which have been converted from Fortran via f2c). As an alternative, Navid has compiled libf2c (precursor to libg2c) from sources on RHEL5. THis seems to do the job. The sources are such that he cannot build a working shareable library with them. This could be fixed, but he would rather not tamper with the sources. For the time being he's built a static library which can be linked into applications needing this support.

  • Windows disk defrag (Navid) has not heard back from SCCS about this. However, builds are now taking 6 hours again (as opposed to 9 or 10) so there probably isn't much to be gained by defragmenting. (Leon) Why is 6 hours typical? Builds of Gleam on his laptop take about 20 minutes. (Navid) The disks on the Windows RM machine are very slow. They always have been slow and apparently nothing can be done about it.

  • SCons and Windows (Joanne) has a workable, if unpleasantly complex, way to produce a set-up file on Windows, which, when run, will set environment variables appropriate for an end-user's or developer's installation. The file can be created by the user at the command line or as an SCons target (it cannot be created as part of an RM build because it isn't portable). Next step is to run the file within GoGui (creating it first, if it doesn't already exist) when the user asks for a terminal window or a Visual Studio session.

previous

minutes index

next

J. Bogart, Last Modified: 01-Jun-2010 15:47:33 -0700