Core Minutes 6/28/2011ScienceTools: (Jim) created a new Likelihood tag to fix a bug in WcsMap2 concerning rebinning. This may also be relevant for the gtdiffrsp crashes discussed last meeting.
(Tom G.) Jim and Joanne made patches to suspicious code reported by Jim's run of ValGrind. The latest attempt, using an optimized + debug (that is, optimized but including debug symbols) build of that code, LATEST 3478, had an 11.3% failure rate.
[Late update: LATEST 3483 has Jim's WcsMap2 fix and one more patch for an unitialized variable. Keep your fingers crossed!] [Still later update after trial runs of 100 and then 200 jobs: seems to be ok now!]
FSSC: (Eric W.) Some users would like to see pointlike added to the FSSC distribution. However pointlike doesn't meet all the requirements for ftools. (Elizabeth) The first big hurdle is to get it to build.
Documentation: (Richard) No real plan yet since Chuck's departure. Karen has agreed to handle the Newsletter and perhaps the LAT Performance page. Tutorials and the Workbook are still up in the air. There will be a meeting tomorrow (Wednesday) at 10 AM on the subject.
Pass8: (Tracy) GR v19r1p0 is out and being exercised; a round of fixing is in progress. There are issues with background generation. flux is generating 5% very high energy (20 TeV) alphas.
The Pass8 workshop is coming up in just 2 weeks; it will have a full schedule.
Tracy has been working on Johan's request for multiple overlay input streams. The original design allowed for only one and it has taken some doing to change that, but the code is now checked in. Johan needs to try it out.
More GR (Heather) There is a new patch of G4 which we'll probably install, but not immediately.
Francesco is still checking out various options for multiple scattering simulatin
There is a new branch tag for Level 1: v17r35p20. It's being checked out now.
Disks: (Richard) The 10 Oracle servers we ordered are beginning to arrive; two are certainly here, perhaps four. We owe six to other groups. 24 Tbytes of pre-launch stuff got nuked, giving us about a 3-month cushion. The full Pass7 Recon is still scheduled for August.
(Heather) u30 has been cleaned up to some extent: it's now down to 87% usage, which will do for the time being.
Windows RM builds (Tom S.) They were stuck again because the known_hosts file had disappeared yet again. He restored it and things are back to their previous state: builds are mostly succeeding except for several test programs which fail to link because the linker can't find the main entry point. This started to happen at the same time we began to explicitly force 32-bit builds on 64-bit machines.
python (Heather, Jim) There have been reports of problems running code using our python distribution where the error message mentions being unable to load md5 (part of matplotlib). Jim reports this may be due to a related issue of hashlib going missing - though we see it (and md5) in our python distribution. Heather is holding out hope that this will be fixed with our next version of python 2.7.1.
In addition, there have been ongoing reports of trouble with pre-installed IPython we started to distribute with python. It turns out that upon installation, the path is hard-coded and is not easily fixed. This obviously causes grief for users installing python outside of GLAST_EXT. The workaround we will pursue is to avoid distributing IPython as part of the python external delivered by the installer. Users will be directed to install IPython on their own (it's a one-liner: "pip install IPython" ) The GLAST_EXT installation of python will have IPython included for SLAC users. [update: Heather has made the new tarballs for use by the Installer as proposed.]
Windows updates (Heather) David Jackson contacted us concerning patches that need to be applied to glast05, glast06 and glast10. glast10 is the only box of those that is used in CMT builds. glast05 is a test box as far as I can see. glast06 used to be part of our build queue, but was having technical difficulties and was removed some months ago. It may be time to decommission glast05 and glast06.
GRBAnalysis (Joanne) wrote a setup script for GRBAnalysis-scons which should meet requirements more or less, but probably could stand some improvements. We'll know better once Nicola et. al. have tried it out.
tagCollector (Joanne) fixed a fatal but straightforward bug (required a one-line change) in tagCollector. The bad version was apparently installed some time after June 20th and the fix went in on the 22nd.
Announcements
SCons meeting Tomorrow (Wednesday), 11:30 AM PDT, following a meeting on documentation at 10.
Farewell, Eduardo His last day is Thursday.
Bon voyage, Heather ..also leaving Thursday, but [I am especially pleased to say, since I will be back-up] returning July 17th.
|
|
minutes index
|
next
|