Choose timezone
Your profile timezone:
Powered by Indico
v3.3.2-pre
Attendees: Danilo, Guilherme, Philippe, Zhe, Jim
Danilo: Working on two test failures. Rare thread-safety issue: when reading TTrees from multiple thread, in some rare case, one of the subbranch is not put in memory correctly (the object address is wrong). A fix with a (possibly too large) extended critical section in the address setting code. We do not understand yet what is the actual problem. Just protecting the GetInfo help almost as much .. but still fails 1 out of 150,000 runs.
Philippe: I recommend ….
Danilo: indeed.
Danilo: The TFileMerger test is failing in the Ofast (fast math) build due to a different compression ratio. I am tracking down if the data change is expected or not.
Danilo: Parallel (tp parallel branch processing) fails (1 out of thousands case), we see a double delete. The TBufferFile that is a member of TBasket is being delete twice. This is odd because there is only one TTree per thread/task.
Philippe: working on two problems, one where a namespace named ‘next’ is confusing rootcling and the other where we see ‘rare’ segmentation fault when executing JITed code.
Zhe: not much too report this week.
Viktor: not much too report this week.
Jim: not much too report this week. Continuing work on the same project … working on generalizing loops for femtocode.
Guilherme: Looking at TMVA failure. Not obvious why the network training is not converging.
Guilherme: Can you remove the vecCore repository so I can move mine into place.
Philippe: Due to the holiday in Europe, let’s skip next week’s meeting.