Core Minutes 7/31/2012

• ScienceTools: (Jim) Just one item: in binned analysis the usage of a keyword introduced recently was inconsistent, causing lack of backwards-compatibility. This is now fixed. See Science Tools Development Notes for details.

• Face-to-face:(Richard) and Jim will be visiting Stephen Fegan and Emilia Becheva at LLR to see about getting her going to help with ST development.

• Hardware: (Tom G.) We'll be replacing the glastlnx machines with 10 new ones. The order is on hold currently, waiting to see if EXO wants to join.

  Upgrade of the nfs servers is in progress. The first of the wains has been drained of xroot data; it will become the user disk. Three more wains are in the process of being drained. We hope to reclaim the space on the old user disk.

  (Richard) We're still looking into the Dell 64-core solution that Stuart Marshall is preparing for the PPA community hardware purchase. This is to replace 400 cores of our batch allocation.

• Pass7 (Leon) Not much news. He's been working on two calibration projects: new calibrations for L & EO and automated calibration creation. Automation looks like it's working for dead strips. (Tom G.) How does automated calibration affect L1? (Leon) The issue is cadence: how to arrange it so we don't have to go back and reprocess. It only takes 4 runs to generate a new dead strips calibration. A couple months of data are required for some of the CAL calibrations.

  For L & EO TKR is done; CAL should be done in a few days.

• Pass8: (Tracy) The new version of Overlay in new tag GR v20r4p1 fixes the problem which had been seen with the SCons build of v20r4p0. One or two job options snafus have seen been found and fixed, the last remaining barrier to MC generation. We'll still with this tag in preparation for collaboration meeting.

  Luca has started the CAL recon reworking as promised.

  We're this close to a runnable vc90 build. test_Gleam runs 600 events, but then crashes. Most of the fixes to code have to do with satisfying the vc90 compiler, which is very fussy about bounds checking. (Leon) Would it be possible to describe the kinds of things you've found either in a presentation or some sort of blog? (Tracy) Certain patterns in our code involving iterators and containers which appear in many places are not acceptable. Meanwhile he also discovered a couple memory leaks, now fixed. [See also discussion of developer environment below.]

• Core talk:

  • rhel4 users (Joanne) We've been asked to reduce the number of users with access to rhel4 as much as possible. Heather has been beating the bushes, looking for people ready to give up access, but some clearly do still need it.

  • External paths (Joanne) Heather and I created external versions without a compiler component in the path for nearly all externals — all except geant, f2c and Fred — which weren't already organized that way. The new versions exist for OSes other than Windows or rhel4. In most cases the new versions make use of sym links so that there are no additional copies of libraries, etc. GlastRelease-scons/containerSettings has been updated to use the new externals, but the new tag of that package has not yet been promoted to HEAD.

  • SCons RM (Tom S.) Running smoothly.

  • Virtual machines (Joanne) made one more bare-bones appliance, this one with CVS and SCons installed. Johan can use it successfully on at least a couple different machines. Luca B. can use it on his laptop but on his desktop machine it gets stuck during boot.

  • vc90/SCons developer environment (Tracy) has run across some ugly counter-intuitive aspects, particularly having to do with the installed headers. Using the SCons-generated project files the install does not get triggered in a natural way when a package other than the one containing the header has source referencing the header. Worse, since there are multiple copies of the header and since, in some circumstances, the "wrong" (installed) one is the one that shows up while debugging in VS, it's easy to modify that one, rebuild, and discover that the modified copy has been overwritten by the old version. (Joanne) is aware that the handling in this situation is poor (and had been vainly hoping that it wouldn't come up much for developers in real life). Given all the differences between VS and SCons, in particular the lack of the "install" concept in VS, she has yet to come up with a way to get all the dependency logic which SCons has into the project files it writes and suspects that, if there is a way, it wouldn't be possible to implement it on a useful time scale. It might be possible to eliminate installed headers for Windows only but she has strong misgivings about that since then the Linux and Windows builds would be fundamentally different. (Tracy) They always are. (Joanne) Not that different. (Tracy) Still votes for getting rid of the installed headers, as a necessary step towards a usable developer environment. (Leon) What about eliminating them for Linux builds as well? (Joanne) There are reasons for doing it this way, primarily I think to speed up compilation. But, in any event, this is what we've got now, recently validated, and I suspect people would not be eager to go through the validation all over again.

    Action items: Joanne will investigate not installing headers on Windows and will also write up suggestions for using the system as it stands.

  • CMT tags restored (Joanne) [should have mentioned this during the meeting but forgot]. The problem of missing CMT tags mentioned last week has been found and fixed. There used to be a cron job running on glastlnx14 every minute which would invoke the oldStyleTagger script. That script looks for new SCons-style tags and makes the corresponding CMT-style tag. The cron job probably stopped running around the time of the glastlnx14 hardware problems. It doesn't show up in the current crontab or trscrontab; John Bartelt found it in an archived crontab for glastlnx14. It appears we are unable now to start a cron job from glastlnx14, but we don't want it running there anyway since that machine will be going away (and although we won't need CMT-style tags any more for rhel4, we may still want them for vc71). Tom S. started it up on glastlnx20 and it's all working again.

previous

minutes index

next

J. Bogart, Last Modified: