[Clas_offline] minutes from OLTWG meeting Nov 12, 2009

[Stepan Stepanyan]

Hi Marco,

I am not sure how to read your email. What do you mean "freeze CVS asap, 
not today".
At the software meeting couple of month ago I though we came to 
agreement that we
will discuss cvs vs svn after some experience will be gained. Several 
groups expressed
interest to test svn/scon (e1-dvcs, eg1-dvcs, frost ...). So far only 
Eugene made an attempt
to use svn and scon and as I can see from his email he succeeded using 
the staff that
was polished by g12 for about a year, but there are still things that 
either missing or not
ready yet. Why not put more efforts to fix every thing and then take an 
action?

Stepan

Marco Ripani wrote:
> Hi,
> need to freeze CVS but asap, not today. Need to first figure out what 
> the best time is, with minimum impact on the other run groups, in 
> particular due to the missing packages.
>
> Marco
>
> Johann Goetz wrote:
>   
>> Thursday, Nov 12, 2009: SCons Build System and CLAS 6 Software
>> 12:00 pm Eastern
>> Cebaf Center  A-110
>> Call in information: Dial 1-877-643-6951
>> International  access: 1-302-607-2017
>> Passcode: 24630562#
>>
>>
>> RPMs, yum, and all that (Maurizio)
>>       * creating RPM (installable pre-compiled packages with dependency
>>         tree) of the CLAS software is a great idea.
>>       * these RPMs can served with a JLab yum server so that
>>         installation on a Fedora or Red Hat (or CentOS) system could be
>>         done with a line like:
>> yum install clas6-reconstruction clas6-generators etc...
>>       * CLAS12 has already implemented this on several flavors of Linux
>>       * 
>>
>> Eugene's Review of SCons Build System
>>       * Generally favourable review. Eugene points out that there are
>>         many packages missing that are needed.
>>       * Many packages will have to copied over directories from CVS. 
>>               * sconstruct files will have to be made.
>>               * gamecock (TOF calib program) has some problems (ROOT/GUI
>>                 related)
>>               * CLASTool should be fairly easy to get to work now that
>>                 ROOT dictionaries can be built properly.
>> From: 	Eugene Pasyuk <pasyuk at jlab.org>
>> To: 	CLAS Offline <clas_offline at jlab.org>
>> Subject: 	[Clas_offline] testing svn/scons
>> Date: 	Wed, 16 Sep 2009 10:29:31 -0400 (07:29 PDT)
>>
>> Hi all,
>>
>> I tried new software tools today on ifarm. Here are some first
>> impressions about this experience.
>>
>> I followed prescription given on the Wiki:
>> http://clasweb.jlab.org/wiki/index.php/Building_CLAS_libraries_with_SCons_on_ifarm
>> It took 20 minutes get the entire source tree source from svn.
>> Then I tried to build. The very first attempt failed. I figured out that
>> the problem was in suggested environment setup file. I modified a couple
>> of lines in it (2 minutes to figure out and fix). After that I made a
>> second attempt to run scons. This time it worked. It took about 3
>> minutes to build all clas libraries and a just few executables by
>> default: 
>>
>> a1c      countbos  g12Anal  gsim_bat  tagraw    view-kaon  view-tag
>> bosdump  g12skim  skim      user_ana  view-st    view-tracks
>>
>> Out of these only few (a1c, gsim_bat, countbos, bosdump, user_ana and
>> tagraw) are of interest to general CLAS user. The rest seem to be g12
>> specific.
>> After doing some archeology in sconstruct files for about a half an hour
>> and making some modifications to them here and there (I have zero
>> knowledge of scons or python at the moment) I managed to build a few
>> more utilities, particularly DC calibration tools trk_mon, trk_mon_lite
>> and dc3. I failed to build start counter calibration code. I was able to
>> build fdump. gflux would not build because of the compiler errors. I
>> wonder how g12 managed to build it in new build scheme from the source
>> in svn. ced would not build because of the compiler errors too.
>> TOF calibration source code is in svn, but it was not modified for scons
>> at all. It was imported as is from cvs with all Makefiles which will not
>> work because of the different tree structure. sc_mon is missing. As was
>> mentioned at the meeting ec timing calibration utilities are not
>> included as several other general purpose utilities. For some reason
>> gsim_int and gpp are not included.
>>
>> To summarize, I succeeded in building most of the libraries and a few
>> executables. Quite a few tools are missing (gsim_int, gpp, some
>> calibration tools). Some programs which  build fine with make would not
>> build with scons (gflux, ced ...?) I don't know if those executables I
>> built are working, I had not had a chance to test them yet. I have not
>> tried to build any of the event generators with scons.
>> By the way, there are only three of them in svn, while we have more in
>> cvs.
>>
>> All in all, svn works just fine, apart from having different directory
>> structure. scons works for some pieces of the software and does not for
>> some
>> others. 
>> Perhaps it is not quite ready for the prime time yet. Some essential
>> utilities are not working yet in new build. 
>>   
>> -Eugene
>>
>>
>> What We Should Do Now
>>       * Hire some undergraduates to move packages from CVS to SVN and
>>         add sconstruct files to build these packages.
>>       * Armand Fraiser was a big help initially with the move to
>>         SVN/SCons and he could continue to do so.
>>       * There is some detailed help information already, it probably
>>         needs a bit of reorganization on the wiki.
>>       * Should we have a list of programs/libraries that should be moved
>>         from CVS to SVN (Armand already created an spreadsheet that has
>>         some of this information. It will be sent to Craig and he will
>>         look into it.
>>       * We need to freeze CVS today.
>>
>> _______________________________________________
>> Clas_offline mailing list
>> Clas_offline at jlab.org
>> https://mailman.jlab.org/mailman/listinfo/clas_offline
>>   
>>     
>
>