Allow disabling cc's ability to run commands via the CC_FORCE_DISABLE environment variable
#1226
Conversation
| { path = "std::env::var_os", reason = "Please use Build::getenv" }, | ||
| { path = "std::env::var", reason = "Please use Build::getenv" }, | ||
| { path = "std::process::Command::new", reason = "Please use crate::command_new()" }, | ||
| { path = "cc::tool::Tool::to_command", reason = "Bypasses `is_disabled()` check. Use try_to_command() instead." }, |
There was a problem hiding this comment.
This is just for internals, obviously users can continue to use Tool::to_command as much as they want.
| Some("cl") => ToolFamily::Msvc { clang_cl: true }, | ||
| _ => ToolFamily::Clang { | ||
| zig_cc: is_zig_cc(&path, cargo_output), | ||
| zig_cc: is_zig_cc(&path, cargo_output).unwrap_or(false), |
There was a problem hiding this comment.
unwrap_or(false) here is a bit janky. I don't think it makes much of a difference what we report when cc is disabled, though.
|
|
||
| fn autodetect_android_compiler(target: &str, gnu: &str, clang: &str) -> String { | ||
| /// Returns true if `cc` has been disabled by `CC_FORCE_DISABLE`. | ||
| pub(crate) fn is_disabled() -> bool { |
There was a problem hiding this comment.
It's unfortunate that this is toplevel and not on Build, but changing everywhere that produces Commands to take a &Build was far too invasive a change. This already is pushing it.
…BLE` environment variable
NobodyXu
left a comment
NobodyXu
left a comment
There was a problem hiding this comment.
How about adding checks to Build::compile_objects?
It is the function responsible for doing the actual compilation, all other functions compile/try_compile/compile_intermediates/try_compile_intermediates all delegate to it.
|
Tbh this does seem a more general problem with the architecture of Cargo's build scripts rather than cc per se. But if using cc as a proxy helps as a mitigation then I'm not against it. |
3 participants
This came up in RustConf as a desire from users/maintainers of 3rd-party build systems, such as Buck2, Bazel, and their ilk.
The idea is that they should be in full control over building all native dependencies, and currently it's (apparently) very annoying and time-consuming to ensure that no
build.rsis compiling/assembling/linking/etc native libraries on it's own. By setting this flag, they can get an error on any crate that attempts this.In terms of the implementation, this was more invasive than I'd like, and I'm open to suggestions about alternative ways of doing this. I failed to locate a single choke-point where returning a
Result::Errwould sufficiently neuter our functionality.I think the approach I've taken is awkward, and would be borderline-unmaintainable were it not for
clippy::disallowed_methods. That said, we do have access to that clippy lint, which I think pushes this over the edge.