Re: multiple include installs for multilibs

[Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>]

> Well... it's doing what the Makefiles tell it to do. It does seem like
> the headers that are the same (all but target-specific) could be
> installed only once...

Yup, I'd agree that it should be done, if at all possible.

>> Are the target-specific headers unique per multilib?  (I don't think
>> so, but thought I'd ask.)  If they are, those would still have to be
>> installed N times.

> I would imagine that, as some multilibs are 32/64 bit, that
> target-specific (and in particular cpu_limits.h) would have to be
> installed on a per-multilib basis.

Yup, I'd think this could be true.  But, I think we have zero support
for such situations.  In fact, there was a time when the AIX port (due
to my stupidity and no one else's) created multiple target-specific
copies of <target-triple>/gthr.h .  At install time, one overrode the
other thus breaking the installed compiler.  David and I changed it so
that the staged copy of gthr.h was the same for all AIX multilibs
since no other solution seemed available given the basic architecture
of libstdc++-v3 and/or its use of automake and/or the way the compiler
finds headers under multilib situations.

> These headers are already clearly defined and marked in
> include/Makefile.am. So it's really just a question of trying to figure
> out how to express this idea in a way that automake can understand.

Surely, automake supports conditional rule-enablement keyed off
whether the Makefile is being built in the primary multilib
configuration directory verses secondary configurations directories.
However, I don't know how to do it offhand.

Regards,
Loren