linux seg fault in ZLibCompressor

[Michael Hunley]

I have an app that runs on a linux webserver, is compiled using gcc 3.1 and 
uses the ZLibCompressor from Crypto++ 5.1 (not current with CVS for several 
months) to compress the data sent to/from the client.  I am getting a 
segmentation fault in the ZLibCompressor::m_adler32::TruncatedFinal(), 
though it seems confused whether it is actually calling the Adler32 or the 
HashTransformation base class (which is pure virtual and would explain a 
seg fault).  I am not that strong in gdb so am not sure how to check the 
vtable to see if it is set correctly in a release build (optimized, but 
including debug symbols).  Can anyone tell me how to check that?

I noticed that HashTransformation does not declare a virtual destructor, 
which I have seen cause problems in other code in the past, and Adler32 
does not declare the overloaded functions as virtual.  That last should not 
matter, but could it be causing gcc to get confused even though the 
signatures are the same?

I've posted a related question several months back concerning the 
ZLibCompressor runtime failure in a completely different app and showing a 
completely different problem. Wei responded that it was most likely a GCC 
optimizer error and the solution was to disable/reduce the optimization 
level.  Given the high volume nature of these apps I cannot just turn off 
optimizations.  Since some of the code is sprinkled through various parts 
of the crypto++ lib *and* a large volume of it is template code from 
headers in my code, how can I selectively turn down optimizations on the 
part that is failing, but keep the rest at higher performance 
levels?  Given the varied nature of the failures, is it really likely that 
it is an optimizer?  Granted, the same code runs fine on Win32.  Has 
anything related to this been fixed in the code base such that trying the 
latest from CVS would be worth trying?

thanks for any help.

Michael Hunley
Senior Engineer
PocketPurchase, Inc.