Re: [3.1-branch PATCH] fix mips CPP size specs for MEABI.

[cgd at broadcom dot com]

At Fri, 26 Apr 2002 11:13:18 -0700, Mark Mitchell wrote:
> > Verified that w/ this patch the compiler & libstdc++-v3 builds for
> > target mipsisa32-elf, host sparc solaris.
> 
> Is this build failure a regression?  (Probably from GCC 2.95?)

mipsisa32-elf is a new target in the last few months.  8-)

this can't really be a regression, but neither can it cause anything
else to regress since it's in a case that's only used by the
mipsisa32-elf target.


> I am relieved to see that your patch only touches systems where O64 is
> the default; those are relatively rare.

s/O64/MEABI/.  Not the same thing.  8-)


> A meta-question: the compiler knows what size_t is, and now the compiler
> and preprocessor are cooperative, so why do we have to put this stuff
> in the specs at all?  I know the answer: stand-alone cpplib.  But that
> just suggests that we should get this information into cpplib, and then
> not have it separately in the compiler.  There's no reason that one half
> of the toolchain can't learn the value from the other half, fixing all
> these bugs once and for all.

I think there's a push (by Neil and others) to move things in that
direction...  but not quite sure where it is at this point.

Really, in general, using specs for ... a lot of things that they're
currently used for (at least in the mips port) is Very Difficult.

For instance, consider the case of:
	-mips3
and (independently)
	-march=foo

for each -march= you need to encode in specs a horrible set of stuff
to get the ISA level right.  It'd made worse (impossible?) if you'd
like to provide predefines which name the currently-selected ISA and
tuning configurations...


cgd