Patch for c/13801

[Joseph S. Myers]

On Tue, 24 Aug 2004, Zack Weinberg wrote:

> I'm sorry, Joseph.  I believe this patch to be based on a design
> decision which is fundamentally incorrect.  I've been sitting on your
> message for two weeks trying to think of something more constructive
> to say, but I can't.

Whereas I believe the design decision (i.e. giving declarations their 
correct types rather than having the datastructures contain wrong data and 
trying to fix it up later in every place it is used) is fundamentally 
correct, and the only alternative you have suggested for how to fix this 
regression is a fundamentally incorrect one of wrong datastructures and 
fixups; that with the patch the compiler can reliably be correct but 
without declarations having correct types it cannot reliably be correct.  
(And I don't like at all the idea of expanding decls with C_DECL_TYPE that 
stores the C type separately from the tree type (to avoid suggested 
middle-end problems) then making every use of TREE_TYPE on expressions or 
decls check for decls and use this type instead; that feels far less 
robust but is the most obvious alternative that would keep all the 
necessary information about types of decls.)  And no-one else has yet 
commented on any of these issues.

-- 
Joseph S. Myers               http://www.srcf.ucam.org/~jsm28/gcc/
    jsm@polyomino.org.uk (personal mail)
    jsm28@gcc.gnu.org (Bugzilla assignments and CCs)