windows-sys v0.59.0 (latest version) hits a FCW on the latest nightly

[tgross35]

I happened to notice this on the playground https://play.rust-lang.org/?version=nightly&mode=debug&edition=2024&gist=ef32dafd88af58c9b698a698fd39d818:

warning: the following packages contain code that will be rejected by a future version of Rust: windows-sys v0.59.0

It appears on nightly but not beta so this must be from a recent change. I don't currently have access to a Windows machine to check out what the error actually is, but considering windows-sys is a top 100 crate, we should probably consider removing the relevant lint from report_in_deps before the beta branch to give some migration time. Otherwise almost everybody developing on Windows is going to have a new error that they can't do much about.

[tgross35]

Cc @kennykerr since I believe you maintain windows-sys

[workingjubilee]

Probably code involving extern "stdcall", cc @RalfJung

[tgross35]

That makes perfect sense, #141435 includes

rust/compiler/rustc_lint_defs/src/builtin.rs

Line 3667 in 1c04750

report_in_deps: true,

. Maybe we could drop the report_in_deps for a version or two so windows-sys has a chance to migrate?

[tgross35]

Linking Zulip discussion https://rust-lang.zulipchat.com/#narrow/stream/122651-general/topic/windows-sys.20FCW

[workingjubilee]

I am revising my guess: it actually is probably usages of extern "cdecl".

[Noratrieb]

I agree that emitting an FCW on every windows build is really bad, we should back that out again. I can do that but will only get to it later today, so if someone else wants to do it that'd be great.

[RalfJung]

Does someone have the actual full error that is being emitted, not just that one-line summary?

Maybe we could drop the report_in_deps for a version or two so windows-sys has a chance to migrate?

Makes sense.

[ChrisDenton]

I'm at a PC now so yes I do: warnings.txt

[tgross35]

To snip one of the messages from the file:

> warning: use of calling convention not supported on this target
>    --> R:\apps\cargo\registry\src\index.crates.io-1949cf8c6b5b557f\windows-targets-0.52.6\src\lib.rs:39:9
>     |
> 36  |   macro_rules! link {
>     |   ----------------- in this expansion of `windows_targets::link!`
> ...
> 39  | /         extern $abi {
> 40  | |             $(#[link_name=$link_name])?
> 41  | |             pub fn $($function)*;
> 42  | |         }
>     | |_________^
>     |
>    ::: R:\apps\cargo\registry\src\index.crates.io-1949cf8c6b5b557f\windows-sys-0.59.0\src\Windows\Wdk\NetworkManagement\Ndis\mod.rs:123:1
>     |
> 123 |   windows_targets::link!("ndis.sys" "cdecl" fn NdisWriteErrorLogEntry(ndisadapterhandle : *const core::ffi::c_void, errorcode : u32, numberoferrorvalues : u32, ...));
>     |   ------------------------------------------------------------------------------------------------------------------------------------------------------------------- in this macro invocation
>     |
>     = warning: this was previously accepted by the compiler but is being phased out; it will become a hard error in a future release!
>     = note: for more information, see issue #137018 <https://github.com/rust-lang/rust/issues/137018>
>     = help: use `extern "C"` instead

Indeed seems to be cdecl. (When somebody updates this lint, probably wouldn't hurt to add the CC name to the warning message - it's a bit hidden in the macro here.)

[workingjubilee]

Opened #142353 to nerf the linting on the relevant fronts for now so we can have this discussion.

[RalfJung]

And that was building for an x86_64 windows target, I assume? Yeah then we would indeed complain about cdecl now.

[jieyouxu]

#142353 merged to downgrade the FCW to not report-in-deps¹, so I believe this is no longer P-critical for prioritization purposes, but the long term discussion of how to transition the FCW to report-in-deps I believe is still outstanding (so this issue should be kept open).

@rustbot label: -P-critical

Footnotes

1. In the short term. ↩

[RalfJung]

the long term discussion of how to transition the FCW to report-in-deps I believe is still outstanding (so this issue should be kept open).

Isn't #137018 the more natural place for that discussion?

[jieyouxu]

Feel free to close this issue in favor of that if that issue is more suitable :) I was just looking at P-criticals