Problems building a profiled ROOT version on SL3

Hi everybody,

on “Scientific Linux SL Release 3.0.3” with gcc323 I tried to build ROOT 4.01.02
with profiling support. I added a ‘-pg’ to the OPTFLAGS in config/Makefile.linux
and started the make. Everything worked as expected up to the point when
rootcint_tmp is run:

Generating dictionary netx/src/G__Netx.cxx…
utils/src/rootcint_tmp -f netx/src/G__Netx.cxx -c -Ixrootd/src/xrootd/src netx/inc/TXAbsNetCommon.h netx/inc/TXConnectionMgr.h netx/inc/TXDebug.
h netx/inc/TXError.h netx/inc/TXInputBuffer.h netx/inc/TXLogConnection.h netx/inc/TXMessage.h netx/inc/TXMutexLocker.h netx/inc/TXNetConn.h netx
/inc/TXNetFile.h netx/inc/TXPhyConnection.h netx/inc/TXSocket.h netx/inc/TXUnsolicitedMsg.h netx/inc/TXUrl.h netx/inc/LinkDef.h
gmake: *** [netx/src/G__Netx.cxx] Segmentation fault

Is this a known problem? Or is there a workaround (or even a solution) for this
problem?

Thanks,
Matthias

Hi,

not seen this problem yet :wink: And no idea on how to fix it since it looks like a compiler issue. You could try profiling with valgrind/cachegrind, which does not need a recompile with a special option (just compilation with the debug option).

Cheers, Fons.

Hi Fons,

thanks for the prompt reply. Running in valgrind has the disadvantage that things
run very slow and it takes ages to profile the BaBar reco and simu applications.
And I guess profiling the code that was compiled with -g will give different results
compared with the code compiled with -O and w/o -g. And we can not simply
compare the outcome with previous releases.

You think it’s a compiler problem. I’ll try with different gcc versions and see what
happens.

Greetings,
Matthias

Hi Fons,

I tried with gcc3.3.1 (instead of gcc3.2.3) and everything worked. Looks like we
had bad luck when choosing the compiler version for BaBar.

Matthias