Open joshtriplett opened 2 years ago
many people have advocated of making the compiler use cc-rs to find the linker
If the suggestion is to introduce dependency on cc-rs, then I for one[!] would advocate against it(*). I reckon that rustc would be better off being self-sufficient and maintaining its own cross-compile prefixing scheme(**) and provide a way for cc-rs to use at its discretion. As opposed the other way around. Just in case, I do recognize the sentiment behind the suggestion to let cc-rs decide over rustc, and even found myself wishing for it to be possible multiple times, yet at the same time I recognize the value of rustc being self-sufficient.
(*) Well, there already is a dependency, on Windows, so I suppose this is an argument against widening the dependency. (**) Yet keep the current way to specify the linker, so that one can override the default for experimental and trouble-shooting purposes.
there already is a dependency, on Windows
For reference. The dependency on Windows is of different nature. cc-rs helps rustc to find the toolchain components it's looking for, but doesn't tell which components to use.
rustc currently has limited knowledge of the correct linker to use on various platforms. cc-rs has a much better idea of which linker to use. There's been a long-standing goal that many people have advocated of making the compiler use cc-rs to find the linker; this would fix cross-compilation for many targets.
Filing this issue to track that possibility. There are also corresponding issues on the rustc repo, which this should link to / be linked from.