[cfe-dev] Assertion on valid use of highlighting whitespace

[Chris Lattner]

r106338 should do this.  Please let me know if you run into any problems with it.

-Chris

On Jun 18, 2010, at 1:26 PM, Douglas Gregor wrote:

> 
> On Jun 17, 2010, at 7:47 PM, Tom Care wrote:
> 
>> I think there might be other things broken in this function too.
>> 
>> I'm having problems trying to get highlighting/fixits working for parts of printf statements (rather than the whole thing).
>> 
>> Currently, if I give a SourceLocation within a printf formatting string, it will highlight it from the beginning until the end of the format string (rather than the end I specified). For example:
>> 
>> /tmp/fixit.c:13:15: warning: precision used with 'n' conversion specifier,
>>       resulting in undefined behavior [-Wformat]
>>   printf("%100.100n", (int*) 0);
>>               ^~~~~
>> 
>> When what I expect is:
>> /tmp/fixit.c:13:15: warning: precision used with 'n' conversion specifier,
>>       resulting in undefined behavior [-Wformat]
>>   printf("%100.100n", (int*) 0);
>>               ^~~~
>> 
>> This ends up breaking the fixits, which delete the rest of the format string rather than the invalid part.
>> 
>> The offending line is lib/Frontend/TextDiagnosticPrinter.cpp:116
>>       // Add in the length of the token, so that we cover multi-char tokens.
>>       EndColNo += Lexer::MeasureTokenLength(End, SM, *LangOpts);
>> 
>> I'm sure this line has importance, but I don't think it is correct.
>> 
>> I'm a bit lost in this section! Does anyone have any ideas?
> 
> The mental model for a source range is that it is a pair of source locations [B, E), where both B and E are expected to point at the beginning of the token. That line above is relexing the token at the location E to find the end of the token. Since you've adjusted the ranges to point at specific characters, when you're getting is E adjusted to the end of the "token" where you're pointing... which is in the middle of a string literal, hence the somewhat odd highlighting behavior.
> 
> We would need to introduce a new notion (e.g., a new "CharacterSourceRange") that forgoes this adjustment and treats the ending source location in the range as character-level positions in the source rather than token-level positions. CharacterSourceRange would only be used in limited places when needed, such as FixItHint and perhaps Diagnostic, and all affected clients would need to be updated.
> 
>   - Doug
> 
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20100618/b4fec661/attachment.html>