Core Minutes 1/20/2004
Present: Pol d'Avezac, Ursula Berthon, Joanne Bogart, Toby Burnett,
Jim Chiang, Seth Digel, Richard Dubois, Dan Flath,
Berrie Giebels, Navid Golpayegani, Traudl Hansl-Kozanecka, Heather Kelly,
Michael Kuss, Matt Langston, Pat
Nolan, Dirk Petry, Sean Robinson, Leon Rochester, Robert Schaefer,
Alex Schlessinger, Tom Stephens, Tracy Usher
-
GlastRelease upgrades: (Toby) New
external libraries (ROOT, Gaudi) have been successfully incorporated into the
production Release Manager build; they're used in the most recent HEAD,
version 222. This version builds successfully on the UW Terminal Server, VS
7.1, with the new CMT (v1r14p..) except for failed compilation in ebfReader
(not required for DC1). Alex has the
new Release Manager running on a RH 9 machine, using the gcc 3.2 compiler
with all the other new stuff. There is one remaining problem having to do
with SLAC-specific configuration, causing test programs depending on xmlUtil
to fail to run. Alex believes he has a work-around. (Richard) Rumor has it
the SLAC-specific configuration causing the problem is a mistake which will
be remedied soon by SCS.
- Science Tools architecture : (James)
- Hope to use Hoops (wrapper for the parameter interface library pil) as
glue to chain science tools together.
- A new package,
tug, (that's Test-as-U-Go) so far just in the users
directory of the CVS repository, contains a couple C++ classes which should
aid developers in keeping their unit tests current with the package code.
Please take a look and send your feedback to James. Pat mentioned the product
DejaGnu, which on first glance appears to be addressing a similar problem,
but is perhaps more akin to our Release Manager.
- The intent of Toby's new
tuple package is to facilitate "row" access to data within an event loop.
Work is in progress to modify Goodi to make it easier to use in combination
with tuple. Feedback on tuple is also solicited.
- Error handling: (Toby, Matt, James)
Toby remarked that exception classes are (properly) finding their way into our
code, but this implies something has to be prepared to catch them. To simplify
this process, he recommends that all exception classes be derived from
std::exception and reimplement its what method, which returns a
character string message. Matt further pointed out that it is convenient for
an individual package to define its own base exception class (which
could be derived from std::exception) so that callers can easily catch just
the exceptions from that package. James mentioned that we might want to
integrate some version of the HEADAS error-handling facility which keeps
per-frame information so that a sort of primitive traceback is available for
crashes even if the code being run has no debug information. [Do we plan to
run code without debug information? ed.] From the RootTalk list Toby has
learned how to turn off ROOT's nasty habit of capturing signals (which tends
to make all errors look like ROOT errors and just confuses debugging). He
plans to put this knowledge to use shortly, with Heather's help.
- RootCnvSvc: (Ursula)
(See also Ursula's
status page.) A version able to convert three TDS objects is about to be
available in CVS (would have been already except for a minor technical
glitch). The conversion service is able to write the objects out from a Gleam
job and also can reconstitute them from ROOT file input. Coming soon: support
for the additional 10 or so TDS objects.
A problem with loading continues to linger. Ursula would like to get rid of
the unpleasant work-around: adding lines to RootCnvSvc requirements to
manipulate LD_LIBRARY_PATH.
How should the files being written be identified? (Richard) Ursula's plan
of using a prefix seems to have the right flavor; this kind of scheme is used
elsewhere.
- CMT release area: (Richard, Traudl)
Last week Richard asked ATLAS folk how they're progressing with using the
release area provided by CMT v1r14. They are fully expecting to use it and
have nearly everything they need (perhaps missing job options?) implemented.
(Traudl) She has a working version which properly handles everything
(executables, libraries, job options, includes, ..). It appears that CMT v1r14
unaided only deals with executables. She had to get patterns, fragments,
etc. from ATLAS to handle other kinds of files which belong in the release
area. So far this work has been done for Linux only. She will put it into CVS
shortly. On Windows, since it has no equivalent of soft links, files must be
copied to the release area, making for a very large installation. (Joanne)
However, only one copy, the one in the release area, needs to be packaged up
for export.
- New Release Manager: (Alex) Should be
ready to turn on the new Release Manager permanently within a day or so. It
includes several bug fixes to avoid the RM stepping on itself. Coming soon:
Windows builds (Richard: our 3 Windows servers are expected in about 10 days)
and use of the batch farm for RH9 builds.
-
OPUS: (Dan)
SLAC Computing insisted our OPUS server be upgraded to RH9,
necessitating an upgrade of OPUS as well, from 4.2 to 4.5. Features we're
making use of include
- ability to start and stop OPUS without a reboot
- support for secure FTP
- ability to run clients on Windows XP
On the immediate to-do list:
- put the code in CVS
- create wrapper scripts around the pipeline code
- interface to the database
- support for recovery after crashes (pick up where processing left off
rather than starting over)
-
Next 6 days: (Richard,
Tom) The remaining data (make that 5 days rather than 6) has been sent to
Tom. He sees only one serious problem with it (DATE-OBS and DATE-END keywords
in the headers are the same in all files, in fact the same as for the first
day of data), but it is one he can fix by hand. He expects the data to be
available from the database in an hour or two.
- Science tools progress:
(Seth, Jim)
- Problems of inadequate input checking previously noted in Likelihood have
been fixed and tagged.
- Support for diffuse emission models exists in flux and is used by
observationSim.
- Try the gui alternative to the command line interface..you'll like it.
Thanks to Pat it now runs on Windows as well as on Linux.
- DC1 report: Come hear Richard talk
about it at the GLAST All-hands meeting tomorrow.
- VS Studio 7.1: Once pre-built external
libraries are made readily available, Windows users are encouraged to move to
Visual Studio 7.1, but there is no expectation of cutting off support for 7.0
any time soon.
- Document effect of MC pruning options (Riccardo) (8/5/2003) It's
documented in the code; Tracy will look into putting the information in a
more accessible place. (9/2/2003)
- Write up, and perhaps present at a future meeting, hints on debugging.
(02/18/03) Some combination of Toby and Tracy will provide
something for Windows.(7/22/03)
- Overhaul documentation altogether by about November (Heather and ??) (9/2/2003)
In progress; not yet done (01/13/04)
- Resolve calibration/timestamp issue. (11/20/02) Online folk have not yet thought about
putting timestamp information into EM data, but, thanks to Richard's
inquiry, may now do so. (2/25/03) We can probably live without a realistic
system for EM, but we will put it on the agenda for the DC1
workshop.(5/6/03) At the DC1 Workshop Steve Ritz promised to find an
authoritative reference for timestamp format and pass it on. (7/17/03) Now
that events have "real" timestamps, Joanne needs to reassess
situation with Calibration (9/2/2003) Timestamp exists in EM data. Needs to
be fetched by Calibration (1/13/04)
J. Bogart Last Modified:
01-Jun-2010 15:45:23 -0700