hfinkel added a comment. In http://reviews.llvm.org/D19678#419445, @rjmccall wrote:
> This discussion of the command line interface makes me think that we should > be taking Richard's suggestion one step further. Why is Clang's involvement > here more than just handing down specific requests for optimization data to > LLVM and packaging that information back into some reasonable format? The static analyzer supports outputting its data in plist format (with which I'm not familiar in detail, but it looks like a fairly-simple xml format). Is that close to what you had in mind? Maybe YAML would be better (since LLVM actually has a parser for that)? > The actual presentation of that data seems like it belongs in a separate > library / tool, which can have a rich set of visualization options. This is > also a more rigorously testable design, since the tool has a well-defined > input format that's not just an incidental by-product of the compiler. I think this makes sense. > If we did that, then Clang just needs (1) an output filename, (2) an optional > list of passes to collect data from, and (3) maybe some stringly-typed > configuration data for each. Sure. Although I'd prefer to leave the filtering to the tool unless the I/O requirements become unmanageable. Users don't know, and shouldn't know, what passes do what. http://reviews.llvm.org/D19678 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
