[cfe-dev] Diagnostic Improvements

[Chris Lattner]

On Nov 17, 2008, at 1:55 PM, Ted Kremenek wrote:
> by passing a FunctionDecl*) in arbitrary ways.  For example, one can  
> imagine that the HTML diagnostics engine (used by the static  
> analyzer, but also usable for regular diagnostics) would add  
> hyperlinks from function names to their definition.  In a similar  
> way IDEs could leverage

That would be cool.

>> because FormatDiagnostic would be given the QualType, so it could do
>> any crazy AST-level groveling that it wants to format it based on  
>> what
>> the DiagClient implementation wants.  This also is a big step to
>> solving #3, because it would mean that we could write things like:
>>
>> DIAG(err_operator_overload_must_be_binary, ERROR,
>>     "overloaded '%0' must be a binary operator (has %u1 parameter
>> %up1)")
>
> I think that in the end we should probably go farther than this and  
> not have any English text in the character string passed to DIAG

Yes, I agree.  We already aim to avoid this.

> since we have to think about localization from the get-go instead of  
> retrofitting it in.  This of course is problematic for custom  
> diagnostics.


At this point, I'm not too worried about custom diags.  They are  
basically an escape hatch.  Lets solve the simple case first :)

-Chris