[PATCH, C] Fix PR40563

[Shujing Zhao]

On 07/27/2010 09:58 PM, Joseph S. Myers wrote:
> On Wed, 21 Jul 2010, Shujing Zhao wrote:
> 
>> Index: c-decl.c
>> ===================================================================
>> --- c-decl.c	(revision 162361)
>> +++ c-decl.c	(working copy)
>> @@ -4103,6 +4103,35 @@ start_decl (struct c_declarator *declara
>>    return tem;
>>  }
>>  
>> +/* Subroutine of finish_decl. Diagnose uninitialized const member of type TYPE.
>> +   DECL is the original declaration. */
> 
> This comment is unclear.  What you seem to mean is: TYPE is the type of an 
> uninitialized object DECL.  If that object has a const member, diagnose 
> this.
> 
Thanks.
> Why do you need to recurse down among fields?  Why isn't checking 
> C_TYPE_FIELDS_READONLY enough?
> 
It used to notice which field should be initialized, just like the diagnostics 
at c++. If the notice is not very necessary, C_TYPE_FIELDS_READONLY is enough.

> What if the original uninitialized object is not a structure but an array 
> of structures?  Will you diagnose things in that case?

C++ would not error an uninitialized array of structure that has a const member. 
  So I think it doesn't need warn at C?
> 
> You don't appear to have any testcases where the recursion is needed for 
> diagnosis, or where the member in question is an array, only tests 
> directly involving a const scalar element of the structure or union.
> 
Yes, they should be added.

Pearly