Rust: gfx_device_gl 0.16.1 broken on latest nightly

[bjorn3]

Caused by #63613. To be precise the https://github.com/rust-lang/rust/pull/63613/files#diff-c915449964a130718bd732c4f035b2cfR101 change.

It basically does:

extern crate gfx_core as core;
format!(...);

Because format! uses ::core::format_args to refer to format_args!, it will try to resolve it in gfx_core, which doen't contain format_args!.

   Compiling gfx_device_gl v0.16.1
error[E0433]: failed to resolve: could not find `format_args` in `core`
   --> /Users/bjorn/.cargo/registry/src/github.com-1ecc6299db9ec823/gfx_device_gl-0.16.1/src/shade.rs:434:24
    |
434 |                 name = format!("Target{}", index);
    |                        ^^^^^^^^^^^^^^^^^^^^^^^^^^ could not find `format_args` in `core`
    |
    = note: this error originates in a macro outside of the current crate (in Nightly builds, run with -Z external-macro-backtrace for more info)

error[E0433]: failed to resolve: could not find `format_args` in `core`
   --> /Users/bjorn/.cargo/registry/src/github.com-1ecc6299db9ec823/gfx_device_gl-0.16.1/src/shade.rs:487:30
    |
487 |             let color_name = format!("Target{}\0", i);
    |                              ^^^^^^^^^^^^^^^^^^^^^^^^ could not find `format_args` in `core`
    |
    = note: this error originates in a macro outside of the current crate (in Nightly builds, run with -Z external-macro-backtrace for more info)

error: aborting due to 2 previous errors

For more information about this error, try `rustc --explain E0433`.

[Centril]

cc @petrochenkov

[jaynus]

This regression affects another project I am working on as well. is there any status on this?

[nikomatsakis]

I'm not sure what the details of #63613 were, but this feels like a bug to me -- in particular it feels like the behavior here is less hygienic, not more? @petrochenkov perhaps you can clarify your view on the situation, that'd be great.

[petrochenkov]

Well, it's more hygienic than before because previously format used whatever format_args was closest in scope at the call site, while now it certainly uses format_args from the crate named core.
(That doesn't make the change less breaking.)

What macro_rules! format really wants is to use def-site hygiene so that "the crate named core" is searched from the macro_rules! format's definition point of view. (Not possible right now due to def-site hygiene for crates not being implemented.)
In the meantime the change to macro_rules! format can be reverted to fix the regression.

[nikomatsakis]

@petrochenkov

Well, it's more hygienic than before because previously format used whatever format_args was closest in scope at the call site, while now it certainly uses format_args from the crate named core.

I see -- just not necessarily the core that was in scope at the point of definition.

In the meantime the change to macro_rules! format can be reverted to fix the regression.

Seems right. Is there a tracking issue to cc or something?

[petrochenkov]

That said, the constructions

extern crate some_crate as core;
extern crate some_crate as std;

or (same, but from command line)

--extern core=/path/to/some_crate
--extern std=/path/to/some_crate

really require some responsibility since they effectively override std and core.

If you override them you should be able to provide all the necessary services expected from std and core, since for any third party macro ::std::foo is the only reliable way to refer to the standard library, at least until def-site hygiene is fully implemented and stabilized.

[petrochenkov]

@nikomatsakis

Is there a tracking issue to cc or something?

This is supposed to be the tracking issue for the regression?

[est31]

While most of the changes in @petrochenkov 's PR have increased hygiene, the specific change linked by @bjorn3 didn't do any increase (or decrease) as both times the macro is being resolved in an unhygienic way.

From a pragmatic point of view, lots of people seem to shadow the core crate but few people want to (saying "want to" and not "do" because of selection bias) shadow the format_args! macro while wanting to use the builtin one for format! which is what the change now allows you to do. Furthermore, not breaking downstream users should always be a priority. It's not their fault that the format macro uses something in an unhygienic way.

IMO the specific change should be reverted.

since for any third party macro ::std::foo is the only reliable way to refer to the standard library, at least until def-site hygiene is fully implemented and stabilized.

There is a pattern to use std or any other third party crate in a hygienic way. You define a #[doc(hidden)] pub mod __reexports { pub use std; } in lib.rs and then use $crate::__reexports::std::... to refer to std items in the macro.

[petrochenkov]

Fixed in https://github.com/rust-lang/rust/pull/64060.

[nikomatsakis]

check-in from compiler triage:

marking as p-high, but @petrochenkov has a pending fix already 🎉