[RFC/objc] Support for 'Class <Protocol>' & frontend cleanups

[David Ayers]

Hello Zem,

Ziemowit Laski wrote:

> Sorry for the delay -- I was really busy with the mainline merge (the 
> ChangeLog was a real pain :-) ). 

No problem...

> Anyway, I think your patch does the right thing, at least in principle. 

Hmm, well they were /two/ patches, and both were actually more WIP than 
finished patch proposals.  I wanted to know which direction I should 
continue.
Option 1. Make 'Class' an ObjC-Keyword (just like 'id')
Option 2. Remove 'id' as an ObjC-Keyword (and thereby simplify c-parse.in)
I'm heavily leaning towards 'Option 2', but I'm lacking the insight on 
the potential interference with ObjC++.

> One thing that would need
> improving is the test cases -- they'll need to be dg-fied (look at any 
> .m file in gcc/testsuite/objc.dg/ to see what I mean).

(I thought I mentioned that...) :-)

>   Oh, and I guess you'll need a ChangeLog entry. :-( 

Of course, once it is posted as an RFA.  And I do plan to split the patch:
1.) white space patch to cleanup in accordance with some GNU coding 
standards.
2.) renaming patch.
3.) 'Class <Protocol>'-Support patch (either 'Option 1' or 'Option 2')
4.) Deprecating '<Protocol>' as a synonym for 'id <Protocol>'
... Then I'll look into the following:
5.) Review handling of error_mark
6.) Introduce 'same_type_p' handling

> So, can you re-spin it against the mainline _after_ the 
> objc-improvements-branch work goes in (which should hopefully be 
> tomorrow)? 

Will do.  Actually I had just finished bootstrapping patch 1.) (which 
turns out not to be a pure white space patch as I introduced a few 
brackets to support the auto indention) and was about to post it, but it 
could wait.  OTOH, It may be easier to merge in now that to merge back 
into the branch later.  It's your call, ping me when you're ready.

Cheers,
David