search

mystor / rust-cpp

Embed C++ directly inside your rust code!
Apache License 2.0
795 stars 44 forks source link

Prefer intrinsics and primitives in core #78

Closed richardeoin closed 4 years ago

richardeoin commented 4 years ago

Rather than the equivalent exports from std

See https://doc.rust-lang.org/core/index.html

Useful for creating no_std crates, and is completely equivalent in other cases. There are still some features that require std, using these in a no_std crate will error at compile time (the current behaviour).

Mark the cpp crate as no_std

ogoffart commented 4 years ago

Thanks or the patch. This looks quite good. However, tere are still some usage of std:: in cpp/src/lib.rs I guess they should be removed before marking the crate as no_std.

ogoffart commented 4 years ago

There is also one use of ::std::sync::Once in the cpp_macro generated code. That needs std. But maybe the crate can still be marked as no_std even if it it stays.

richardeoin commented 4 years ago

The no_std meta in cpp/src/lib.rs actually applies to almost nothing. The macros are expanded in the user's crate, and it doesn't propagate into the cpp_macros crate because that is marked with:

[lib]
proc_macro = true

But still they can be replaced with core at no cost, I will do that.

As you say there's no trivial replacement for ::std::sync::Once, but this is only occurs if expanded into the user's crate (through the use of the rust! macro, I think). So much of this crates functionality can be used by no_std crates anyhow.

ogoffart commented 4 years ago

Thank you !

ratijas commented 3 years ago

I'd like to note that while the PR is good in itself and mostly works in the wild, it was released under patch version bump despite being a breaking change (https://github.com/woboq/qmetaobject-rs/issues/108) — which is bad.

Incompatibility arise from the fact that in Rust 2015 edition using ::core paths require explicit extern crate core; line at the top of the main.rs file, while ::std paths work out-of-the-box.

I probably should open a new issue with a request for updated documentation regarding 2015 edition, or write such PR myself...