Patch: Remove non-mapped location support

[Tom Tromey]

>>>>> "Manuel" == Manuel López-Ibáñez <lopezibanez@gmail.com> writes:

Manuel> What else would be required to have caret diagnostics? We
Manuel> could add the option to 4.4 under -fcaret-diagnostics
Manuel> (following LLVM).

The front ends have to be fixed in a couple ways.

First, they have to always use the correct location when emitting an
error.  I fixed the C front end to do this a while back, but I did not
start on C++ (see my change from 2007-11-08 to see what I did).

Second, they have to always use the correct location when constructing
any tree that includes a location.  I have not started on this at all.

Ideally we would get rid of input_location entirely.  There's no need
to have a global variable for this kind of thing.  Whether or not we
get there is another question.

I'd like to get column numbers working but I have a lot of stuff going
on... so if you want to help with that, that would be awesome :)

Tom