Patch: Remove non-mapped location support

[Richard Guenther]

On Mon, Feb 25, 2008 at 5:40 PM, Tom Tromey <tromey@redhat.com> wrote:
> In 4.3 we changed the default location representation to "mapped
>  locations".  The old representation was deprecated at that point.
>
>  This patch removes the old, deprecated location code.  By and large
>  this is a mechanical patch.  There are a few details worth mentioning:
>
>  * I poisoned 'USE_MAPPED_LOCATION' to prevent regressions due to
>   merges.  I think there's no point in poisoning the other deleted
>   input.h things; a merge would run into other errors if those are
>   used.
>
>  * java/java-tree.h still mentions USE_MAPPED_LOCATION in a comment.
>   This is a reminder to me to look into this -- the comment is for a
>   field which is apparently never set.  I will fix this separately.
>
>  * input.h used to claim that location_t was a deprecated typedef.  I
>   think that removing this is not very important and I "un-deprecated"
>   it.
>
>  * The Makefile.in change is needed to make gengtype work properly now
>   that I've deleted the input.h-related special case.
>
>  * There are a few places that had odd comments related to mapped
>   locations, e.g. look at cp/decl2.c.  I left the "???" comments
>   (slightly modified) for language maintainers to look at.  They don't
>   seem to cause any problems with the test suite -- and anyway 4.3 is
>   already running this branch of the conditionals...
>
>  * tree-vectorizer.h defines shorthand for various input.h-defined
>   things.  In general I think this is not the best thing to do; it is
>   better, in my view, for all the code in GCC to use a single API,
>   rather than having modules defining idiosyncractic variants.  I did
>   not change this.
>
>  Bootstrapped and regtested C, C++, ObjC, ObjC++, Ada, Fortran, Java,
>  and Treelang on x86 FC-6.  Ok?

The middle-end parts are ok.

Thanks,
Richard.