Re: OBJ and EXE and cross-compilation (Was: AC_OBJEXT again)

[Mo DeJong]

On 12 Dec 2000, Akim Demaille wrote:

> I don't think we need the test we used to perform (compiling and
> checking for the names the compiler leaves around).  But if we want to
> run this test, then only when not cross-compiling.
> 
> There are other things which might be considered.  After all, when
> cross-compiling, how about checking $host to compute CYGWIN32 etc?
> This would relieve us from having to compile something.  Sure it is
> not `clean', but anyway we are already using tables for this case.
> There's no way to be sure of the extensions used on the host platform,
> since we'd need to run something on it.

If I was compiling with a Linux cross mingw compiler, then
that would work, since I would pass --host=i386-mingw32msvc
to the configure script, it would find i386-mingw32msvc-gcc.

But would this still work if I wanted to compile with
the mingw compiler under Cygwin? The mingw native
compiler is also named gcc, so what --host triple
would I need to pass to the configure script? It
seems like I would need to pass something for the
--host argument, otherwise the configure script
would not think it was a cross compile. Is this
even a cross compile? Well sort of. (ugh)

If I did not pass a host triple, would the
configure script detect Cygwin even though
CC was set to the mingw compiler? That
seems like the logical thing to do if we
only use uname. Problem is, if configure
sets a bunch of Cygwin #defines and then
tries to exec the mingw compiler,
all hell will break loose.

Mo DeJong
Red Hat Inc