Re: [Boost-users] [Boost.Multi-Index] ICE w/ VC7.1

From: Marc Viala (mviala_at_[hidden])
Date: 2007-01-03 17:21:26

De la part de John Maddock
> Marc Viala wrote:
> > Before trying to recompile some codes from B.MI/test to crash my
> > compiler w/ an ICE, I've just played with some compilation options
in
> > my IDE and the results are:
> >
> > In Debug mode:
> > If the C/C++ options "/Zi" & "/Gm" are set: KO -> w/ ICE in
> > apply_wrap.hpp
> > If the C/C++ options "/ZI" & "/Gm" are set: KO -> w/ ICE in
> > apply_wrap.hpp
> > If the C/C++ option "/Gm" is not set and "Zi" or "ZI" is set: OK
> >
> >
> > In Release mode:
> > with or without the "/Gm" option set: OK
> >
> > Funny... It seems that the /Gm option is quite unstable in debug
> > mode? I don't know exactly the goal of this option but our project
> > template set this option by default in this mode.
> >
> > Joaquin or John, do you know the meaning of this compilation option
> > for VC7.1? Do you know the recommended options w/ this compiler to
> > use Boost libraries, more specifically B.MI? For information, up to
> > now, we use "/ZI" or "/Zi" in conjunction with "/Gm" option with all
> > other Boost libraries.
> 
> With apologies I think I misread your original post: I took your post
to
> mean you were seeing a runtime crash, but you mean an internal
compiler
> error: and I did see that at one point when testing your original
code.  I
> use the same project options as you BTW /Zi and /Gm: the /Zi enables
build
> and continue, /Gm enables a minimal rebuild (compile only the changes
to a
> source).  Generally they work quite well, but if you see an internal
> compiler error the trick is just to do a clean, then rebuild, and that
> usually clears the problem.  Of course cleaning and rebuilding a large
> project can wipe out the advantages that /Zi /Gm normally bring :-(

Thanks a lot this explanation... It does clarify for me the
signification of these options. Now that I'm aware of /Gm option
behaviour I would pay more attention.

> 
> Strangely I was unable to reproduce the internal error this morning no
> matter options I used: it seems to be very dependent on whatever VC
has in
> it's cache.
> 
Yes, that's a pity...

I can now go ahead with B.MI.

Best regards

Marc