Implement tool_lints

[flip1995]

cc #2955

Waiting for rust-lang/rust#52851

The (ui-)tests need to be adapted to the tool lints, by renaming every clippy_lint to clippy::clippy_lint. Also #![feature(tool_lints)] needs to be added to the tests, until tool_lints are stable.

[flip1995]

FIXMEFIXED: I get a weird error here:

error: proc-macro derive panicked
 --> tests/run-pass/used_underscore_binding_macro.rs:7:10
  |
7 | #[derive(Deserialize)]
  |          ^^^^^^^^^^^
  |
  = help: message: called `Result::unwrap()` on an `Err` value: "failed to parse derive input: \"/// Test that we do not lint for unused underscores in a `MacroAttribute`\\n/// expansion\\n#[deny(clippy::used_underscore_binding)]\\nstruct MacroAttributesTest {\\n    _foo: u32,\\n}\""

That's because somehow syn fails to parse the new tool_lints...
See serde and syn

[oli-obk]

we just need to update our dependency version requirement to the most recent one, then it'll work

[oli-obk]

Why do we even need this imported here? Shouldn't declare_clippy_lint be enough? Maybe we need to import it in the expansion of declare_clippy_lint

[flip1995]

Because we use the declare_clippy_lint! which expands to ...declare_(tool_)lint!....

Maybe we need to import it in the expansion of declare_clippy_lint

That would probably be cleaner.

[flip1995]

Or not: If we use declare_clippy_lint multiple times in one file we would get multiple use rustc::declare_tool_lint. And that happens pretty often.

[oli-obk]

can we just call rustc::declare_tool_lint! instead of importing it?

[flip1995]

That should work. I'll try this.

[flip1995]

If we use rustc::declare_tool_lint! in the macro_rule, we have to import use rustc in each file, or else:

error[E0433]: failed to resolve. Use of undeclared type or module `rustc`
  --> clippy_lints/src/lib.rs:24:9
   |
24 |           rustc::declare_tool_lint! { pub clippy::$name, Warn, $description, report_in_external_macro: true }
   |           ^^^^^ Use of undeclared type or module `rustc`
   |
  ::: clippy_lints/src/bit_mask.rs:87:1

[oli-obk]

:( ok, lets leave it as is then and punt to the future

[Manishearth]

Does rustc allow you to use the old way of specifying these?

I feel like we should have a week or two where rustc will match both clippy::foo and foo and warn on the latter. Or we can register deprecated aliases or something here. Not sure how to best make this work.

[flip1995]

You can still use the declare_lint! macro and nothing would change. If we switch to the declare_tool_lint! macro the old ones won't work anymore. That is because I implemented this macro to just concat the tool name onto the lint name. This means there is no good way for rustc to allow both IMO.

I could have added a double check for tool_name::lint_name and just lint_name to the compiler and remove it in several weeks again. But then most of the handling of the tool_attributes and tool_lints should be done by the tools and not the compiler, like the RFC stated.

My solution to this would be to write a lint for occurrences of #[cfg_attr(feature="cargo-clippy", ...) before making this change (And maybe wait a week or two).

We also need a lint for unknown_clippy_lints in the near future, after landing this.

...I should list this in #2955

[flip1995]

Whoops my update script for the ui-tests slipped a little bit there. I'll reduce the number of lines changed a bit 😄

[Manishearth]

@flip1995 I meant that we can introduce such a warning into the rustc codebase itself, yeah? "If you use a lint that shares a name with a tool lint, here's a warning"

[Manishearth]

for tool lint groups we can temporarily make the registration API ask for a deprecated fallback name

We're the only consumers of this API, we can make changes like this 😄

[flip1995]

I think I finished the ui-tests update. I looked through every .stderr file and github reports +2,243 −2,243 for the two stderr commits. 🎉

[flip1995]

Yes, we could lint in the rustc codebase for lints that are from tools aka Clippy. But this could only detect unscoped lints from Clippy if we compile the code with Clippy in the first place. In that case also Clippy can lint for usage of the unscoped lint names. (I'm sorry, if I'm missing the point of doing this in the compiler. Doing the ui-tests update was pretty annoying exhausting 😄)

for tool lint groups we can temporarily make the registration API ask for a deprecated fallback name

That makes sense to me.

[Manishearth]

But this could only detect unscoped lints from Clippy if we compile the code with Clippy in the first place. In that case also Clippy can lint for usage of the unscoped lint names.

Yes, this is only if you're running with clippy.

(I'm sorry, if I'm missing the point of doing this in the compiler. Doing the ui-tests update was pretty annoying exhausting smile)

Doing this in the compiler lets us have the lint allow still work. Otherwise the clippy user will get the deprecation warnings, as well as tons of unknown lint warnings and tons of silenced clippy warnings that get unsilenced.

[flip1995]

Ahh, I got it now! Thanks for the clarification!

[flip1995]

Travis should be green with the next nightly now. Do not merge this until the backwards compatibility is implemented in rustc.

[phansch]

@flip1995 looks like some conflicts appeared

[flip1995]

Yeah, I currently working on the backwards compatibility of the tool_lints. After that I'll have to make adjustments here anyway. Also it will need a rebase. But I want to do this all at once.

[oli-obk]

ping @flip1995 any update on the backwards compat?

[flip1995]

I will open a PR in rust-lang/rust by the end of this week

[flip1995]

It's almost finished, just need to prettify it a little. Sorry for the delay.

[Manishearth]

Rebased and pushed, and pinned daa4f0a to the tool-lints branch on this repo. When rust-lang/rust#53762 lands, please ensure that that commit gets into clippy master via git merge (NOT rebase)

[Manishearth]

rust-lang/rust#53762 landed. I'll do a local test run and catch any merge issues that may have happened