[Halld-offline] Fix for mcsmear crashes

[Sean Dobbs]

Hi Offliners,

I've just checked in a fix for the mcsmear crashes that some of you were
seeing that resulted in truncated mcsmear output. Some details of the
problem and its fix are given below.  Please let me know if you see any
more similar problems.


Details:

Back in last October, during the first detcom simulation run, it was
noticed that many jobs were crashing at the mcsmear stage, with a truncated
output file.  This behavior was noticed by others later on (e.g.
https://halldweb1.jlab.org/mantisbt/view.php?id=424 ).  David Lawrence
identified the proximate cause as due to a string field of
DBCALSiPMSpectrum objects being read in with a length of -1.  In HDDM I/O,
string lengths are treated as unsigned ints, so this resulted in the
program assuming an extremely long string length, which overflowed buffers
all over the place.

I checked that the hdgeant output files were uncorrupted and didn't contain
the -1 length values, but was eventually able to determine that the problem
came from reading in very long strings from the HDDM files generated by
hdgeant.  Figuring out the ultimate cause was very difficult because (1)
there is no error checking done for input/output on uncompressed HDDM
files, (2) whatever error apparently overwrote memory in a difficult to
diagnose way, so it was difficult to get any information out of the istream
objects.

Changing the size of the input buffer used to read in the files didn't fix
the problem, so instead I changed the coding of the DBCALSiPMSpectrum
string to store consecutive 0 values as "X<N>", where <N> is the number of
zeros, since the long strings were dominated by these strings of zeros.  In
some limited testing, this seems to have fixed the mcsmear crashes, and
should provide a mild reduction in size of the output files.

I think that this problem is peculiar to the output of HDGeant 3, since I
believe it is the only program which still produces uncompressed HDDM files.


Cheers,
Sean

-- 
Sean Dobbs
Department of Physics & Astronomy
Northwestern University
phone: 847-467-2826
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailman.jlab.org/pipermail/halld-offline/attachments/20150115/ad9ec1f1/attachment.html