richard.t.jones at uconn.edu
Wed Jan 30 15:54:16 EST 2013
Just a brief update on this from my side. I devoted about 1 week of work to this back in November, planning to finish the first G4 release by the end of the year. If you recall, this was my plan stated at the June collaboration meeting.
After I got 1 week into it, a great thunderstorm of updates to the geant3 tree burst upon me, obsoleting a good porton of my work, and I immediately halted development. Since then, it has all pretty much gone stale, and I would have to go back to the beginning. The reason these two are coupled is that the upgrade entails wrapping the G3 hits code into classes that fit the G4 interfaces.
As you may imagine, that entails changes to the existing hits code structure, although not its behavior. When I wrote the original hdgeant hits code, I tried to be careful about what was exposed and what was not so that I could do this transition smoothly, but as you may imagine, many cooks have been working in the kitchen since then.
I am willing to go back and start it up again, but we cannot have ongoing development of the G3 code happening at the same time. I simply cannot keep up. The G3 code is the reference against which I am forming the hdgeant4 library classes. If you can convince everyone to freeze the trunk G3 code, I can get back to work on the upgrade. What nobody wants is a release of G4 that effectively rolls back everybody's simulation updates since who-knows-when.
On 1/30/2013 1:34 PM, Matthew Shepherd wrote:
> Hi software gurus,
> Many years ago we decided that migrating to Geant4 was not a high priority for our software activities, although our software was intentionally designed to facilitate a migration at a later date. With the generation of the first data challenge now under our belt, I wonder if we want to revisit this decision?
> I think there are two main reasons to consider making a switch:
> Running CERNLIB on modern operating systems is a real headache. I know there are working solutions that allow us to function, but they work on only some OS'es and involve downloading patched code a private developer. CERNLIB has no official support and overall this results in a net productivity loss for us since we depend on it and it generates a lot of frustration for people trying to get our software running.
> The second, more significant reason is that, if we are going to switch to Geant4, we want to have fully made the transition to Geant4 by the time we start taking data. One of the most time consuming and challenging aspects of analyses is understanding the systematic uncertainties in efficiency due to the fact that the Monte Carlo isn't a perfect model of the detector. We'd like to only go through this once: we don't want to do it once for Geant3 and then again for Geant4. Geant3 is fine for the exercises we are doing now, but if we're ever going to make the switch, it would be wise to do it well before we start taking data.
> Of course it is always easy for someone to propose work when that person is not the one actually doing the work! The burden would fall on the Geant expert (or experts, do we have more than one?) in our group. I know that Richard reported last year that significant progress had been made in porting code to Geant4.
> Maybe we can bring this up for discussion at a future software meeting or discuss more through this email thread. Maybe the right answer is that it is still not a high priority item -- I think it merits a little discussion again though. (Maybe you already talked about it during one of the many software meetings I missed last fall!)
> Halld-offline mailing list
> Halld-offline at jlab.org
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 3232 bytes
Desc: S/MIME Cryptographic Signature
More information about the Halld-offline