Present: Joanne Bogart, Anders Borgland, Toby Burnett, Jim Chiang, Johann Cohen-Tanugi, Seth Digel, Richard Dubois, Warren Focke, Navid Golpayegani, Traudl Hansl-Kozanecka, Heather Kelly, Michael Kuss, Matt Langston, Julie McEnery, Pat Nolan, James Peachey, Leon Rochester, Bob Schaefer, Tom Stephens, Tracy Usher
Core issues: (Toby) See a summary of a few of this week's headliners, including the Top Ten (packages with most compilation warnings) and discussion of how to deal with paths in cmt requirements files. Toby noted that several Release Manager builds are hung; did Navid's code to detect this and send out notification work as expected? (Navid) Yes, he has several emails and presumably Alex (not present) does as well. [A couple additional issues from Toby's list are discussed at greater length below.]
Random, Valgrind: (Michael) See his page on all the work he's done to date investigating reproducibility of events. In an attempt to track down the last per cent or less of unexplained non-reproducible events, he tried Valgrind, a tool for detecting memory problems which "we" (well, Navid) have used before, though not for a while. It detected several uninitialized variables. At least one was significant.Initializing it properly has reduced the number of non-reproducible events and may have fixed other problems as well. A couple observations on the use of Valgrind:
(Richard) We should think about integrating use of Valgrind into Release Manager tasks. Can we make a system test to keep tabs on the reproducibility? (Michael) Need to have runs with different sources and, in some cases, different choice of algorithms; perhaps this is better done as a unit test, since package owners know what choices are appropriate. His current method for determining reproducibility is to enable certain kinds of logging and then compare (long) logs. (Toby) Should suffice to just output state of random # generator at end of each event. (Michael) Yes, you would think so, but he has observed counter-examples.
On the Windows front, we're getting closer to being able to dump 7.0. Everyone at SSC has the new compiler, though many are not yet using it. Toby has purposely kept the Glast-ts terminal server at 7.0, but expects to upgrade it to 7.1 when he returns to Seattle.
Science Tools: (James) Work with Yasushi to convert EventBin to use tip, etc., has gone well, a pair-programming success story. Need to find a new name. This might be a good time to discuss naming conventions altogether for Science Tools. Preference expressed for all lowercase.
JIRA: Johann is still learning about it. For the most part it appears to be pretty easy to deal with. Would like to implement feature allowing JIRA read access to our CVS repository. That way it could "know" about bug fixes to particular reported bugs. Matt notes that our evaluation copy is only good for another 48 hours.
Installers: The current version Pat is playing with must be invoked interactively (no scripting), so could not be used by the Release Manager. Do we want to spend several $K on something which can be?
Tape archive tool: (Navid) It's ready to go, except fro a grouping-by-run-number feature. (Richard) this should take the pressure off our limited disk space for the time being.
New Root macros for analysis: (Heather) Motivated by some suggestions from Matt on current Root best practice, she's rationalized the RootTreeAnalysis macros by using Selectors, Folders, Tasks and tree friends. It's all nicely described in this page.
Flight Integration: The weekly run-down:
previous | minutes index | next |
J. Bogart Last Modified: 01-Jun-2010 15:46:26 -0700