[WIP] Expand all attributes in left-to-right order

[petrochenkov]

Fixes #83331.
r? @Aaron1011

[petrochenkov]

@bors try

[bors]

⌛ Trying commit 6a42bf1e0b4588b67e16e436325a62eecd8e5f78 with merge 99a443d6903f71d521237d8acb5a3016b534d17b...

[bors]

☀️ Try build successful - checks-actions
Build commit: 99a443d6903f71d521237d8acb5a3016b534d17b (99a443d6903f71d521237d8acb5a3016b534d17b)

[petrochenkov]

@craterbot check

[craterbot]

👌 Experiment pr-83354 created and queued.
🤖 Automatically detected try build 99a443d6903f71d521237d8acb5a3016b534d17b
⚠️ Try build based on commit f826641, but latest commit is 6a42bf1e0b4588b67e16e436325a62eecd8e5f78. Did you forget to make a new try build?
🔍 You can check out the queue and this experiment's details.

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🚧 Experiment pr-83354 is now running

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[Mark-Simulacrum]

@craterbot retry-report

[craterbot]

🛠️ Generation of the report for pr-83354 queued again.

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🎉 Experiment pr-83354 is completed!
📊 73 regressed and 16 fixed (152359 total)
📰 Open the full report.

⚠️ If you notice any spurious failure please add them to the blacklist!
ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[petrochenkov]

There's a variety of errors (some of them are strange and I need to check why they happen in more detail), but my impression so far is that we won't be able to do this.
There are too many uses like

#[generate_something]
#[cfg(FALSE)]
ITEM

where "something" is unintentionally generated despite the cfg(FALSE), and is producing errors.

We should be able to report warnings for out-of-order expansions though (similarly to #79202).

I haven't seen regressions from #[cfg(TRUE)]s being removed after expansion (they were previously left in place), but, again, I need to look in more detail. Perhaps we'll be able to land at least this part.

[petrochenkov]

Analysis of regressions.

Group 1: Proc macro attribute is written before #[cfg(FALSE)] so it either doesn't resolve due to missing imports or items introduced by the same cfg, or generates some code referring to the ITEM assuming it exists.
Resolved by simple attribute reordering.

#[wasm_bindgen] #[cfg(FALSE)] ITEM ++
#[wasm_bindgen_test] #[cfg(FALSE)] ITEM +
#[bench] #[cfg(FALSE)] ITEM +++++++++++
#[global_allocator] #[cfg(FALSE)] ITEM +++
#[ctor] #[cfg(FALSE)] ITEM +
#[tokio::test] #[cfg(feature = "tokio_support")] ITEM +
#[tokio::test] #[cfg(feature = "test")] ITEM +
#[derive(FromPrimitive)] #[cfg(feature = "serde-serialize")] ITEM +
#[derive(MessageEventContent)] #[cfg(feature = "unstable-pre-spec")] ITEM +
#[pin_project] #[cfg(feature = "discover")] ITEM +

Group 2: #[cfg_attr] disables the legacy derive helper resolution heuristic (#79202) so #[serde] no longer resolves.
Resolved by simple attribute reordering.

#[serde] #[cfg_attr(TRUE, derive(Serialize))] ITEM +

Group 3: #[test_case] (https://crates.io/crates/test-case) processes other #[test_case] attributes in its input specially, and #[cfg_attr]s hide them from it.
Cannot be resolved by simple attribute reordering, needs #[cfg_eval].

#[test_case(args)] #[cfg_attr(TRUE, test_case(args))] ++

Group 4: Some lints use item.attrs.is_empty() as a heuristic, so they start working differently if #[cfg(TRUE)] is removed (#84110).

#[cfg(TRUE)] { EXPR } // unnecessary braces +
#[cfg(TRUE)] extern crate foo; // `extern crate` is not idiomatic ++

[petrochenkov]

Blocked on #87220.

[bors]

☔ The latest upstream changes (presumably #84956) made this pull request unmergeable. Please resolve the merge conflicts.

[petrochenkov]

Superseded by #92473.