? or % can't be used with o!(), looks like regression

[netvl]

It seems that the logic in this PR: #170 has been reverted somehow, because I now get exactly those errors which are described in that PR when I try to use ? or % with o!():

error[E0277]: `*mut (dyn std::ops::Fn() + 'static)` cannot be shared between threads safely
   --> src/server/mail/watcher.rs:755:20
    |
755 |         logger.new(o!("part_path" => ?p))
    |                    ^^^^^^^^^^^^^^^^^^^^^ `*mut (dyn std::ops::Fn() + 'static)` cannot be shared between threads safely
    |
    = help: within `[std::fmt::ArgumentV1<'_>]`, the trait `std::marker::Sync` is not implemented for `*mut (dyn std::ops::Fn() + 'static)`
    = note: required because it appears within the type `std::marker::PhantomData<*mut (dyn std::ops::Fn() + 'static)>`
    = note: required because it appears within the type `core::fmt::Void`
    = note: required because it appears within the type `&core::fmt::Void`
    = note: required because it appears within the type `std::fmt::ArgumentV1<'_>`
    = note: required because it appears within the type `[std::fmt::ArgumentV1<'_>]`
    = note: required because of the requirements on the impl of `std::marker::Send` for `&[std::fmt::ArgumentV1<'_>]`
    = note: required because it appears within the type `std::fmt::Arguments<'_>`
    = note: required because it appears within the type `slog::SingleKV<std::fmt::Arguments<'_>>`
    = note: required because it appears within the type `(slog::SingleKV<std::fmt::Arguments<'_>>, ())`
    = note: required because of the requirements on the impl of `slog::SendSyncRefUnwindSafeKV` for `(slog::SingleKV<std::fmt::Arguments<'_>>, ())`
    = note: required by `slog::OwnedKV`
    = note: this error originates in a macro outside of the current crate (in Nightly builds, run with -Z external-macro-backtrace for more info)

And indeed, the kv! macro now uses __slog_builtin(@format_args) instead of whatever is implemented in #170, which in turn implements the pre-#170 approach of using format_args.

[dpc]

I thought there was a test and all. Can you check which version of slog are you using exactly?

I'd appreciate some help. Not sure when I'll be able to dig into this.

[nazar-pc]

2.5.2 all the way down to 2.4.0 has this problem.
2.3.3 failed with some other error, so I'm not sure about it.

[dcormier]

Still an issue on 2.7.0.

[panicbit]

It looks like the original fix got reverted in 4695580 and fcf3efe.
This also got documented in the changelog.

I would be glad to fix this issue again, but it would be helpful to know why the change was considered broken back then. Otherwise I feel like the risk of wasting my time is too great.

[dpc]

🤷 No idea, sorry. (but PR with some tests for it would be awesome.)

[ppershing]

It looks like the original fix got reverted in 4695580 and fcf3efe. This also got documented in the changelog.

I would be glad to fix this issue again, but it would be helpful to know why the change was considered broken back then. Otherwise I feel like the risk of wasting my time is too great.

I believe this is the relevant report of breakage: #176 (comment)

[Techcable]

In addition to fixing this, we're going to have to add a bunch of tests so this never happens again.

I've considered switching to a procedural macro, but that would likely require depending on syn.