c99 VLA semantics

[Mike Stump]

On Jun 16, 2006, at 5:44 PM, Mark Mitchell wrote:
>> I haven't tested the patch and it was submitted without testcases,  
>> but it
>> ought to fix
>>
>> int a;
>> struct s { void (*f)(int (*)[a]); };

I'm happy to include your testcase in the submission, thanks.

> Mike, please resubmit the patch, with test cases, and with analysis of
> why the other uses of the function do not need to change.  Please
> mention that this is a regression when you resubmit.  If you CC: me, I
> will review.

This fixes a regression in my recent VLA work caused by a pre- 
existing latent bug in tree.c.  I'v analyzed all the uses of  
var_mod_type_p and variably_modified_type_p in gcc/{*,*/*,*/*/*} and  
I think all uses by the 4 C language frontends are ok, I think  
treelang, Fortran and Java are ok, Ada I am unsure about and have  
sent a pointer to an Ada person.  I have reason to think they should  
explicitly review and ok the change.  The only use by the non- 
language frontends that might concern me is the tree inlining code.   
I'm unsure of the impact this change would have on it.

The testcase is Joseph's work.

bootstraped and passed reg testing on powerpc-apple-darwin8.5.0 for -- 
enable-la
nguages=c,c++,objc,obj-c++.

Ok?

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: vla-4ntc.diffs.txt
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20060617/2908d0e4/attachment.txt>