Re: configure problem for cross compilers

["Jan Beulich" <JBeulich at novell dot com>]

>> Trying to build an x86_64 compiler (3.4.0) on an x86 machine, I ran
into what seem to be a general problem with the top level configure
>> script forcing LD to be set (for libtool) but that way confusing the
gcc/configure script's determination of what ld to use (namely does
this
>> prevent a newly built ld in the same tree to be used). This used to
not be a problem until 3.3.3.
>> Is this intentional (and thus there is a known solution * other than
having to pass in LD=ld each time one wants to run a cross configure
>> * that I simply wasn't able to locate), or should I file this as a
bug? In the latter case, I'm also unclear where such a bug for the
top-level
>> stuff should go (as this is not gcc-specific).
>
>It isn't clear to me what problem you are trying to reprort.  Please 
>give more details.  We can't help unless we can reproduce the
problem.
>
>It looks like you are complaining abou the gcc_prog_ld code in the 
>toplevel configure.in, but I don't understand why it is a problem.

Yes, correct.

>You say this was never a problem until gcc-3.3, but it looks like this

>code is new in gcc-3.4, so if there was a problem with gcc-3.3, it
must 
>have been a different one.

I may have expessed this ambiguously: It was no problem up to and
including 3.3.3; I started seeing the problem only with 3.4.0.

>This sets LD which is supposed to be the native LD which goes with CC.

>This shouldn't be a newly built linker, particularly if you are trying

>to do a cross build.

That would be my understanding, too. However, gcc/configure.ac at line
1691 (3.4.0 released code) uses LD (as well as AS a few lines earlier)
to determine the linker to use for the target. I would call this
incorrect (it should use LD_FOR_TARGET here instead, at least for cross
builds), but since this was there in 3.3.3 (and I can track it back to
at least 3.2.0) it may be that there are other (external) dependencies
on this (e.g. [RPM] build scripts of third parties) and that this was
intentional, which is why I asked the question here in the first place
rather than strait going into bugzilla.

>It is OK to report bugs in the toplevel files against gcc.

Thanks, Jan