3 ms·
I'd take a compiler backend that compiles large parts of the modern software ecosystem every day and has countless programmers using it over some niche compiler
by adrianN 7y ago
I'd take a compiler backend that compiles large parts of the modern software ecosystem every day and has countless programmers using it over some niche compiler that is used only for a vanishing fraction of all applications but boasts some kind of official stamp of applicability in safety relevant contexts.
- dodobirdlord 7y agoI'm not so sure I would. For one, Ada is in no way a niche language. For two, C (and C++ by extension) simply were not designed with safety or reliability in mind. I happen to think that it's fair to say that they weren't designed at all, in that C compilers existed before the first language specification, and the language specification was forced to accommodate their quirks. Cargo culting about C is fine and all, but it often leads to absurd claims that it is more suitable in any arbitrary domain than languages and compilers designed specifically for the domain. Behold, LLVM and GCC both miscompile the same numerical code involving mildly tricky aliasing constraints. https://github.com/rust-lang/rust/issues/54878#issuecomment-429578187 https://github.com/rust-lang/rust/issues/54878#issuecomment-... I think I'd prefer that safety-critical code be written in languages that don't allow pointer arithmetic except in scenarios that can be proved via static analysis not to introduce multiple memory aliases. Expecting program writers and compiler authors to get these sorts of things right in C/C++ is just unreasonable.
- adrianN 7y agoYou don't have to use C or C++ to use LLVM. I'd argue that you would probably never find a bug such as that in a certified compiler, not because of the certification, but because far, far fewer eyes look at its output. I have worked in safety critical software and used a certified compiler. It had enough bugs that I personally found four in my first year.