Re: libstdc++ bootstrap failures on sparc-sun-solaris2.8 (analyzed)

[David Edelsohn <dje at watson dot ibm dot com>]

>>>>> Gabriel Dos Reis writes:

Gabriel> Hmm, David fixed a similar issues for AIX with 

Gabriel> 2001-07-23  David Edelsohn  <edelsohn@gnu.org>

Gabriel> * include/bits/limits_generic.h (int): Set digits and digits10
Gabriel> appropriately for word size.
Gabriel> (unsigned int,long,unsigned long): Likewise.

Gabriel> Seems we ought to generalize his approcah...

	I only fixed he infrastructure so that limits_generic.h is correct
for various word sizes.  We still have not made any changes to allow that
correct infrastructure to be utilized for 32x64 multilibs.

	As I said in the previous message, the two approaches I have come
up with are:

1) Always fall back to generic if gen-num-limits fails.
2) Handle the entire multilib as cross-compiling.

Gaby> We're going to try to run a 64-bit app on a system running
Gaby> 32-bit, can't we predict it will fail? If so, why should we try to run
Gaby> it in the first place?

	We can -- sort of.  It is not always easy to determine that we are
running on a 32-bit system that cannot handle 64-bit apps.  I can create a
32-bit app on AIX to find it out, but not just testing some existing file
or flag.

	But what do you to do with that knowledge?  Treating the entire V3
32x64 multilib configure and build as cross-compiling is very complicated
(see the Linux stuff -- we would need that for every target).  The rest of
configure works except for gen-num-limits.  We only want the limits
generation to occur as if cross-compiling.  I do not yet have a good idea
of how or where to have only gen-num-limits care about 32x64.

David