[cfe-commits] r101610 - in /cfe/trunk: include/clang/Basic/DiagnosticSemaKinds.td lib/Sema/Sema.h lib/Sema/SemaChecking.cpp test/Sema/builtin-prefetch.c test/Sema/builtin-stackaddress.c test/Sema/builtins.c

Chris Lattner clattner at apple.com
Sun Apr 18 10:55:42 PDT 2010


On Apr 18, 2010, at 2:50 AM, Eric Christopher wrote:

>>> 
>> 
>> Thanks for doing this Eric.  One stylistic thing: there is no need to say which argument number is problematic, just point to the argument and underline the problematic expression with a source range.  Removing the argument # makes it more concise.
>> 
>> IOW, please change "argument 0 to '__builtin_return_address' must be a constant integer" to "argument to '__builtin_return_address' must be a constant integer" or better yet, "argument must be a constant integer" if its obvious what the callee is.
> 
> Sure, wasn't sure how much parsing of the decls we were supporting vs the underlining :)
> 
> That said the main reasons I threw the function name in there were:
> 
> a) most of the messages already had them,
> b) the most typical case runs like this:
> 
> "some builtin isn't being called with the correct args :
>    _mm_my_thingy(0, 2)
> 
> instantiated from here:
> #define _mm_my_thingy(a, b) __builtin_my_thingy((a),(b))"
> 
> or close enough :)
> 
> So, whatcha think? Leave it in or take it out? :)

Leaving the function name in makes sense, please remove the arg # though, thanks!

-Chris



More information about the cfe-commits mailing list