> Then I don't see how the flag is handled in this patch. Err, we probably don't need to support it here, but the rest of my previous comment is still valid: why don't we perform the has_color check inside changeColor/resetColor. http://llvm-reviews.chandlerc.com/D127