Re: PATCH RFC: Recognize gold in libstdc++ configure script

[Benjamin Kosnik <bkoz at redhat dot com>]

> The libstdc++ configure script does some direct comparisons of the
> linker version number.  These comparisons do not work for gold, a new
> ELF linker.

Yeah. Usually the version number compares are just because we cannot
figure out a better way to do it. (ie, feature test.)
 
> It would not be appropriate for gold to simply pretend that it is GNU
> ld.  Not only is that not correct, it would actually give the
> incorrect result.  The libstdc++ configure script uses a version
> number test to check whether the linker supports --gc-sections; gold
> does not.

OK. From what I understand, gold also does not support things like -z
relo, right? (And the existing configure test doesn't have to be
modified? Seems like it will work as-is.)

So, your behavior is

with_gnu_ld == yes for ld and gold
glibcxx_ld_is_gold == yes for gold only

Seems sane.

I would imagine that eventually you'll have to add something like
with_gnu_ld_gold to top level, and that would subsume
glibcxx_ld_is_gold, huh? Just curious about your longer-term
plan. (I think libgfortran also uses -z relo so it would need
modification to work on too.)
 
> This patch fixes the problems for me.  Does this seem reasonable, or
> would people prefer a different approach?

This seems reasonable to me. Thanks!

-benjamin