[cfe-dev] Objective-C top-level constructs code generation

[David Chisnall]

Hey Daniel,

On 12 Aug 2008, at 01:29, Daniel Dunbar wrote:

> Hi David,
>
>> Sadly, I haven't had a chance to do much with this lately.  I still
>> have a huge pile of clang diffs sitting on my machine waiting to be
>> submitted, and I've made a number of bug fixes in the copy of
>> <snip>
>
>
> As you may have noticed I have started work on support for the NeXT
> runtime. While I'm not going to go out of my way to change the
> CGObjCRuntime interface if there are things that need to be changed or
> improvements to the API I was planning on doing them.

I haven't updated clang for a while, since Chris's changes to the  
ObjC stuff filled my tree with conflicts and after spending two days  
failing to untangle them I ran out of time to work on clang for a bit.

> Obviously this is an issue if you have outstanding diffs vs the  
> current
> interface. I don't want to make extra work for you but its not  
> clear what
> else we can do.

My original plan was to have the runtime-specific code in clang  
isolated and reusable for implementing other dynamic languages on top  
of an Objective-C runtime.  This was possible with the old interface,  
which isolated clang-specific and runtime-specific code in separate  
files.  The new interface breaks this layering, and so I will  
probably have to use the version I have forked for Smalltalk as a  
base for this (the Nu guys and some people working on JavaScript and  
Io implementations contacted me a while ago about using this,  
although I think they also wanted a NeXT runtime implementations  
which is likely to be a low priority for me) and back-port the  
changes to clang periodically.

> What kinds of changes are in the "huge pile" of diffs? Are they  
> primarily
> changes to the internals or have you also significantly changed the  
> interface?
> If it is only the internals then this might not be so bad, but if  
> the interface
> has changed as well would it make sense to simply update the  
> current interface
> to match your revisions so that the diffs will go easier in the  
> future?

I haven't made any changes to the runtime interface - I just haven't  
updated my copy to correspond to the new one.  Before Chris's  
changes, I had ObjC support in my local tree up to the point where  
the only problems compiling GNUstep's Foundation and AppKit libraries  
came from bits of C not implemented.  The Étoilé runtime  
implementation is now only in my local tree - Chris didn't update it  
to his new interface and so the version in trunk became too bit- 
rotted to remain and was removed a few weeks ago.

I have sent most of the changes to the list at various points, but  
haven't had a chance to address all of the review comments in them.   
I don't think many were related to the runtime-specific components.   
There were quite a few little changes related to mapping the ObjC  
type system to LLVM, and a few minor changes in various places where  
I encountered problems compiling ObjC code.

If you have more time to tidy and polish the changes than me then I'd  
be happy to send you a diff from my tree.  I think it was down to  
just over 1000 lines before Chris's changes (not including the code  
to support the Étoilé runtime).

David