10 ms·
From the perspective of a compiler writer, the most interesting thing about this code is that it contains no bounds checks while still being safe. Using the hig
by pcwalton 12y ago
From the perspective of a compiler writer, the most interesting thing about this code is that it contains no bounds checks while still being safe. Using the high-level iteration APIs like map/filter/etc. is not only good for readability, it's good for performance as well. C-style for loops require bounds checks to maintain safety, but there's no need for them with a properly designed iterator API.
- tritri 12y agoThat's actually very interesting, I've had to implement bounds checking in a C-style language, and it definitely was a lot of fun and required some thorough checks. What's taken into consideration when designing the iterator API? Could you please provide an example of using a properly designed iterator API? Thanks!
- andrewflnr 12y agoWhat about all those calls to unwrap? Don't those sweep some of the bounds checking under the rug? It's okay in a proof of concept like this, but I wouldn't be comfortable running that code "for real". Edit: to be more clear, I wouldn't expect it to have an impact on performance, but I'm not convinced this is a good example of expressiveness and safety in the same package.
- mbrubeck 12y agoThe `unwrap` calls are all in the input parsing code (`slurf_file` fails if input is invalid), except for one on the result of `min_by` (`classify` fails if you pass it an empty array). I wouldn't say any of these "sweep bounds checking under the rug."
- ben0x539 12y ago.unwrap() is checked, if it goes wrong, it will safely cause task failure and stack unwinding. There won't be any memory unsafety.
- andrewflnr 12y agoYou can't get corruption, but I don't consider a program that violently terminates on unexpected input to be "safe". Just semantics, I guess.
- dbaupp 12y agoYeah, saying 'safe' around Rust is generally interpreted as meaning memory safety.
- dbaupp 12y agoThe unwraps are not memory unsafe, unlike avoiding bounds checks. In any case, rewriting to avoid them isn't so hard: http://huonw.github.io/2014/06/11/error-handling-in-rust-knn-case-study.html http://huonw.github.io/2014/06/11/error-handling-in-rust-knn... (Notably, since it's all in the types, you're guaranteed to not ignore any errors.)