Re: [objc-improvements] Yet another set of fixes committed

[Ziemowit Laski]

On Wednesday, Aug 27, 2003, at 18:39 US/Pacific, Alexander Malmberg 
wrote:

> Ziemowit Laski wrote:
> > On Wednesday, Aug 27, 2003, at 17:44 US/Pacific, Alexander Malmberg
> > wrote:
> >
> > > Ziemowit Laski wrote:
> > > [snip]
> > > > Please test away!  I will now sync up with HEAD and, if there are no
> > > > regressions, will commit that as well.
> > >
> > > GNUstep fails to build; the first error was caused by a minor problem
> > > in
> > > the new code. However, testing the message passing behavior shows 
> > > that
> > > the new code gets several cases wrong. The commit message implies 
> > > that
> > > there is/was something amiss with the commit. Is this the case? If 
> > > not,
> > > I'll get to work on a patch to clean things up and a set of test 
> > > cases
> > > that covers all the cases I can find.
> >
> > Can you be more specific as to what is broken?
>
> It won't find prototypes for a type if the prototype comes from a
> protocol implemented by the class, eg.:
>
> @protocol Zot
> -(void) zot;
> @end
>
> @interface Foo : Object <Zot>
> @end
>
> int main()
> {
>         Foo *f=nil;
>         [f zot]; /* warns that f might not respond to -zot here */
>         return 0;
> }
>
> I've attached a patch.
>
> Also, it fails to detect multiple applicable prototypes in many cases,
> and the behavior in case d. does not seem to have changed (so it's
> compatible with the old behavior, but it matches neither my nor your
> proposal for new behavior). Anyway, I want to investigate a bit more 
> and
> finish a complete set of tests before reporting on this, so that'll 
> have
> to wait until tomorrow.

Yes, the implementation does depart from both you and I were describing.
Specifically, if your receiver is of type 'id<Foo, Bar, Baz>', then the
front-end will examine the three protocols in turn, and return the first
matching method, rather than search all 3 protocols unconditionally to
reveal any conflicts.  And yes, this is what the compiler did up until
now, and I'm a bit nervous about changing it. :-(  What's your take on
this?

> > Certainly, you're more
> > than welcome to make a patch :-), but perhaps it's something really
> > trivial that I omitted/comitted...
> >
> > Also/more importantly, could you provide me with a quickie tarball of
> > GNUStep sources that I could just run 'make' on?  The GNUStep
> > installation
> > liturgy I found on the Web is highly discouraging. :-(
>
> Installation is easy _iff_ you have all the dependencies covered. 
> Either
> get the cvs version:
>
> http://savannah.gnu.org/cvs/?group=gnustep
>
> (the gnustep/core/ module), or grab a snapshot:
>
> http://ftp.gnustep.org/pub/daily-snapshots/
>
> Running compile-all in core/ should compile and install it. The script
> also shows how to do it manually (in case of errors, or if you want to
> use a special CC or something). Basically, you run "./configure ; make 
> ;
> make install" in make, base, gui, and back in that order.
>
> If you don't have all the dependencies, the build guide is supposed to
> be good:
>
> http://documents.made-it.com/GNUstep/buildguide.html

Yeah, that's the write-up that frightened me so. :-(  That's why I was
hoping that if someone already went through this paint on a RH Linux 8
system, they could just tar up the fruits of their labor, so that I may
jump right into the configure/make step you're describing.  But perhaps
I'm hoping for too much... :-)

> - Alexander MalmbergIndex: objc-act.c
> ===================================================================
> RCS file: /cvsroot/gcc/gcc/gcc/objc/objc-act.c,v
> retrieving revision 1.179.2.7
> diff -u -r1.179.2.7 objc-act.c
> --- objc-act.c  27 Aug 2003 00:19:24 -0000      1.179.2.7
> +++ objc-act.c  28 Aug 2003 01:23:48 -0000
> @@ -6387,7 +6387,7 @@
>        if (CLASS_PROTOCOL_LIST (inter))
>         {
>           if ((meth = (lookup_method_in_protocol_list
> -                      (CLASS_PROTOCOL_LIST (inter), ident, 1))))
> +                      (CLASS_PROTOCOL_LIST (inter), ident, is_class))))
>             return meth;
>         }

Indeed - a stupid omission on my part; many thanks, I'll commit your 
patch,
along with the test case at the top.
--------------------------------------------------------------
Ziemowit Laski                 1 Infinite Loop, MS 301-2K
Mac OS X Compiler Group        Cupertino, CA USA  95014-2083
Apple Computer, Inc.           +1.408.974.6229  Fax .5477