[cfe-commits] r64504 - in /cfe/trunk: include/clang/AST/ include/clang/Basic/ lib/AST/ lib/Analysis/ lib/CodeGen/ lib/Sema/ test/Analysis/ test/CodeGen/ test/Sema/

[Douglas Gregor]

On Feb 14, 2009, at 8:19 PM, Daniel Dunbar wrote:

> On Sat, Feb 14, 2009 at 5:43 PM, Chris Lattner <clattner at apple.com>  
> wrote:
>>
>> On Feb 14, 2009, at 11:32 AM, Daniel Dunbar wrote:
>>
>>>>
>>>> I guess it is bad that we don't accept some programs we used to,  
>>>> but is
>>>> this
>>>> really such a bad thing?  If this just affects stuff in llvm- 
>>>> test, I'm
>>>> inclined to just fix the programs.  If this occurs in some large  
>>>> C code
>>>> bases, then perhaps we should try to be more lenient.
>>>
>>> I'm just concerned about the number of such programs. It seems like
>>> there might be a good tradeoff in allowing some limited but common
>>> patterns of redefinition which are "safe", with the same diagnostic
>>> perhaps but as a warning. This would still put pressure on the user
>>> but at least allow the code to compile. This seems relatively easy  
>>> for
>>> us to do, what is the downside?
>>
>> The problem is that clients of the AST can't make the assumption  
>> that a
>> builtin with a specified ID has the expected prototype.  GCC  
>> doesn't provide
>> this guarantee so all sorts of code has to check that printf has at  
>> least 1
>> argument etc.  It is really annoying and leads to crash-on-invalid  
>> bugs.
>
> Sorry, I guess I wasn't clear.
>
> I wasn't saying we should allow the invalid prototype. I think we
> should just treat it as a warning instead of an error in some cases
> (maybe even all), and the effect will be that we just ignore the
> declaration in the code. I'm certainly not proposing we create
> ill-formed AST nodes.

What if the invalid declaration is also a definition? Do we just skip  
over the entire body?

	- Doug