[patch] Important comment update to gcc/configure.in

[Nathanael Nerode]

Daniel Jacobowitz wrote:
> So let's try this from the top then: if you do not want to assume
> BUILD->TARGET has the same features as HOST->TARGET how do you intend
> to infer the features of HOST->TARGET

I'm not saying you shouldn't do that under some circumstances; I'm 
saying it's worth being very clear about *when* you're doing it.

In order of priority:
1. Manual specification by the user, either as edits to config.h, 
defines on the Make command line, or configure arguments.
2. When the HOST->TARGET assembler is being built at the same time, its 
configuration will tell the features.
3. When the HOST->TARGET assembler *is* the BUILD->TARGET assembler, 
clearly we can test it.
4. If it's different but HOST is a subset of BUILD (host=i386-foo and 
build=i686-foo, for example), we can run the HOST->TARGET assembler 
directly.  (This is probably more trouble than it's worth, so can be 
dropped.)
5. When the HOST->TARGET assembler isn't being built and is unrunnable, 
we can default to the features of the BUILD->TARGET assembler, if you 
like.  It would be safer to assume minimal features.

Either way, the user should be able to override this if she knows 
better, without changing the detected behavior of the BUILD->TARGET 
assembler.  I don't see how that can be done in the current confounded 
setup.