Re: Severe performance problem and proposed solution

[Tom Tromey]

>>>>> "Ralf" == Ralf Corsepius <address@hidden> writes:

Ralf> In its final consequence it would mean either gcc's
Ralf> configuration to be basically broken or autoconf not to be
Ralf> applicable inside of gcc's source-tree, because autoconf
Ralf> pre-requires a functional c-compiler.

I haven't been following this too closely, so I'll just butt in
ignorantly.

Target libraries in the gcc tree are a weird special case.  You have
to do hacks to write their configure scripts.  As long as some hack is
possible, I'm happy.

Removing autoconf from the picture wouldn't be pleasant.  It would
basically mean writing configure by hand.  So let's rule that option
out.

As an example of the hacks, for libgcj we make a decision.  If we're
doing a native build, we freely run link tests and the like.  If we're
cross-compiling (which is a poor heuristic to see if we're targeting
some relatively-OS-deprived box), then we make all kinds of
assumptions about what is available.  This might be the best you can
do.  We could refine how this decision gets made, that would be nice.

Tom