slow V3 configures (was Re: ICE during bootstrap.)

[Robert Lipe]

> > We're pretty certain that those all need to be there, I think.  Whenever we
> > start making assumption about "if foo() is there, surely bar() is also
> > there," it breaks on somebody's platform.
> 
> Right. It's a pain, but necessary. We try to cache most of it, but there 
> are places where it could be speeded up.

If the cache hits were as quick as they are elsewhere in the GCC build
process, they wouldn't stand out.  I know this is pretty subjective,
but on, say, the libiberty or gcc directory when we go through the
"checking for XXX (cached)" steps, they scroll by quickly enough as to
be unreadable and are therefore pretty much unnoticable.  The v3 config
process goes through hundreds of them and takes about a banana and a
half ("one banana", "two banana", ...) per entry.  So a cache hit is
taking on the order of second and a half.  If this was a one-time cost,
it'd sure feel better.

This is on a 400Mhz P-II with fast SCSI drives; not exactly state of
the art, but generally not a horrible place to do GCC development.
(It's not like I'm on a Sparc Classic or something.) Is it this painful
everywhere or am I just lucky in some way?

Perhaps I'm oversensitized becuase I've been chasing problems in the
v3 configure/build process and have thus had to sit through this
many-minute process more times than I feel I've deserved to. :-)

RJL