Core Minutes 11/20/2002  

Present:  Joanne Bogart, Toby Burnett,   Johann Cohen-Tanugi/Michael Kuss, Marco Frailis,  Traudl Hansl-Kozanecka, Heather Kelly,  Sean Robinson, Leon Rochester, Alex Schlessinger, Tracy Usher, Karl Young

• Release management: (Traudl) Traudl presented her proposal on release management and package organization. See her presentation and still more extensive documentation on the subject since the brief summary included here cannot do justice to all the material covered.

  The presentation began with a list of requirements and assumptions, mostly familiar from the way we already operate. However, responsibility would be more clearly laid out in the new scheme: one person would be responsible for the release as a whole, but would depend on other individuals responsible for specific domains. Another new feature of this system would be the ability to use private versions of some packages (on your own local machine) with the standard versions of others located at SLAC via AFS.

  See the file GleamPkgDependencies-update (.ps, .ppt) for diagrams of the proposed organization into containers. Container packages specify complete tags in their requirements files for constituent packages. Leaf packages specify only reduced tags, such as v2r*. A special container package will be devoted to external library interface packages.

  The proposed actions include 

  • creating the new container packages with suitable names
  • several forms of clean-up within requirements files
  • clean-up of CVS repository (e.g., actually deleting from repository obsolete and useless packages now in the attic)
  • more radical reworking of repository. Make a new repository and import active packages in such a way that their relationship within the repository reflects the container package hierarchy. Include mechanism to make it easy to switch between old and new repository.

  Joanne raised objections to a couple bits of the proposal:

  • One of the proposed forms of requirements file clean-up is to eliminate "redundant" use statements. For example, since the xml package uses facilities, packages using xml need not also explicitly use facilities. However, eliminating such statements leaves us with a relatively fragile system where knowledge about what really uses what (as opposed to what requirements files say) has to be more widely distributed among developers.
  • The  second stage of changes to the repository ties physical representation to logical relationships. Since it's awkard to change the physical representation, we loose some flexibility in changing the logical relationships.

See also Toby's Core Issues/Status document for some comments on several of the items below.

• Releases: (Toby) In spite of last week's statement that v3r2 would be the end of the road for Gleam, we're up to v3r4 and still counting. We don't yet have anything to replace the Gleam checkout package. At some point as we switch to the new tools mainline development will become incompatible with existing Gleam releases. By then we need to apply a branch tag to a stable Gleam release (i.e., apply it to the used versions of all referenced packages). We still need to work out details of how one continues to do development, presumably just bug fixes, on the release branch.

• Tools status: (Toby) See this page devoted to Gaudi v11r1 issues. Short summary: looking good so far, at least on Windows with Visual Studio 6.0. It would be nice to be able to treat Gaudi as an external package, but we still need the ability to step through the source in the debugger. Have yet to come up with a scheme for Gaudi as external package which provides this debugging ability.

  The new CMT, v1r12p20020606, is the default on the Terminal Server, but, although available on SLAC Linux, is not yet the default and is (probably -- difficult to be sure) not being used by the nightly builds. Marco had used it successfully on Linux, however there was one small glitch. When the new CMT was checked out some files which needed to be executable had only the read bit set, so Marco had to adjust the protection bits of these files by hand, via chmod.

• Use of stamps: (Marco) Changes Marco made to GlastPolicy  should eliminate problems people have had on Linux with builds not always building everything they should.   For a complete explanation of how this works, and what developers need to do in various circumstances, see the latest version of GlastPolicy mainpage.h.[or, a doxygen version]

• ACD system tests:  (Heather) Heather and Analia have written several ACD system tests. Work is still ongoing. They could be incorporated in the Gleam release now but there is no necessity to do so. 

• Calibration and timestamps: (Joanne) In order for the calibration conversion service to work properly, it has to be possible to translate the timestamp information in a given event to a calendar time. In the very short term we don't need  realistic time values, though it would save trouble later if the translation mechanism were moderately realistic. As soon as someone wants to be able to use the service with data from a prototype instrument we will need realistic time values as well.

  Action Items 

• Respond to Traudl's release management proposal.(11/20/02)
• Come up with procedures for development within release branch (11/20/02)
• Resolve calibration/timestamp issue. (11/20/02)
• Update Gleam User's Guide. (9/25) (10/30/02) Mostly done.
• Assemble minutes, etc. from Core Week reviews of AcdRecon and GLAST Gaudi use. (9/18/02)
• Alex will look into technical solutions (front-end script?) for controlling CVS commit access. (6/5/02) Doable; also circumventable. A back-burner project. (7/10/02)
• CAL geometry changes (5/8/02) Partly done. (7/3/02) Awaiting further information from CAL (7/24/02)

previous

minutes index

next

J. Bogart Last Modified: 04-Aug-2004 15:41:56 -0700