[cgma-dev] cgm-cubit on 32 bit machine

[Jason Kraftcheck]

On 07/01/2011 11:00 AM, Iulian Grindeanu wrote:

>> 
> So the corruption in that string happens exactly when I step over the
> NCubitFile::CCubitFile ccf; statement.
> 
> That class/object has something, because I cannot step into any of its
> methods (or constructor), the debugger skips over it; I will try
> rebuilding with -O0 flag (it is configured with enable-debug and
> disable-optimize, but I do not see -O0 anywhere in makefiles)
> 
> Does anyone else, besides Rajeev and I, builds cgm on 32 bit machine?
> 

If you're linking against Cubit, then your choice of compiler flags will not
help.  Cubit is already compiled, with whatever compiler flags the Cubit
developers used.

The problem is almost certainly that the CCubitFile class increased in size
between Cubit 12.0 and Cubit 12.2.  The header we have is from Cubit 12.0.
So the compiler allocates enough space on the stack for the 12.0-sized
CCubitFile object but the actual constructor in libcubiti19.so initializes
members of a 12.2-sized CCubitFile object, which happens to be bigger and
therefore overwrites other stuff on the stack.

The best solution to this, if possible, is to get a copy of the CCubitFile
source from Cubit12.2 and update our copy appropriately (possibly with some
preprocessor logic if we want to stay compatible with other versions of Cubit.)

If that is not possible the next best thing would be to guess a sufficiently
large size (with enough time and some memory tricks, we could probably
narrow that down fairly well) and add some junk (e.g. a char array) to our
version of CCubitFile to make it sufficiently large.  Of course this is
unlikely work work if CCubitFile has inlined functions for which the
implementation or the location of the relevant data in the class changed.

- jason