Reduce explcit clones

[MichaReiser]

This PR removes some explicit clone in the code base by:

• Deriving Copy for field-less enums and pass them by value. This won't improve performance but signals to readers that copying is cheap.
• Avoid cloning scopes and parents during the deferred checking phase. Cloning isn't necessary because scopes and parents are append-only, and truncate is sufficient to restore them to their previous state.

I used this opportunity to align the deferred checks and replaced the

self.deferred.list.reverse();

while let Some((...)) = self.deferred.list.pop() {
	...
}

with

while !self.deferred.list.is_empty() {
	let list = std::mem::take(&mut self.deferred.list);
	
	for item in list {
		...
	}
}

This has the advantage that we avoid reversing the list (O(n)) and fixes the potential bug where an item added to self.deferred.list during traversal is processed before the other deferred items (because reverse is called before the loop)

CPython Benchmark

The performance improves a little.

hyperfine --warmup 10 \
             "./ruff-main ./crates/ruff/resources/test/cpython/ --no-cache -e --select=ALL" \
             "./target/release/ruff ./crates/ruff/resources/test/cpython/ --no-cache -e --select=ALL"
Benchmark 1: ./ruff-main ./crates/ruff/resources/test/cpython/ --no-cache -e --select=ALL
  Time (mean ± σ):     786.2 ms ±  14.6 ms    [User: 7788.2 ms, System: 306.2 ms]
  Range (min … max):   766.3 ms … 811.4 ms    10 runs
 
Benchmark 2: ./target/release/ruff ./crates/ruff/resources/test/cpython/ --no-cache -e --select=ALL
  Time (mean ± σ):     775.4 ms ±  14.0 ms    [User: 7784.1 ms, System: 274.0 ms]
  Range (min … max):   755.4 ms … 801.5 ms    10 runs
 
Summary
  './target/release/ruff ./crates/ruff/resources/test/cpython/ --no-cache -e --select=ALL' ran
    1.01 ± 0.03 times faster than './ruff-main ./crates/ruff/resources/test/cpython/ --no-cache -e --select=ALL'

[MichaReiser]

Current dependencies on/for this PR:

• main
  • PR Reduce explcit clones #3793 👈

This comment was auto-generated by Graphite.

[github-actions]

PR Check Results

Ecosystem

✅ ecosystem check detected no changes.

Benchmark

Linux

group                                      main                                   pr
-----                                      ----                                   --
linter/all-rules/large/dataset.py          1.00     14.6±0.02ms     2.8 MB/sec    1.01     14.7±0.03ms     2.8 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.00      3.8±0.01ms     4.4 MB/sec    1.00      3.8±0.00ms     4.4 MB/sec
linter/all-rules/numpy/globals.py          1.00    411.9±1.86µs     7.2 MB/sec    1.02    418.1±6.34µs     7.1 MB/sec
linter/all-rules/pydantic/types.py         1.00      6.3±0.01ms     4.0 MB/sec    1.00      6.4±0.01ms     4.0 MB/sec
linter/default-rules/large/dataset.py      1.00      7.8±0.01ms     5.2 MB/sec    1.00      7.8±0.01ms     5.2 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.00   1726.5±4.18µs     9.6 MB/sec    1.00   1722.5±2.33µs     9.7 MB/sec
linter/default-rules/numpy/globals.py      1.00    181.3±0.38µs    16.3 MB/sec    1.01    183.3±0.95µs    16.1 MB/sec
linter/default-rules/pydantic/types.py     1.00      3.6±0.01ms     7.1 MB/sec    1.00      3.6±0.01ms     7.0 MB/sec

Windows

group                                      main                                   pr
-----                                      ----                                   --
linter/all-rules/large/dataset.py          1.00     15.6±0.07ms     2.6 MB/sec    1.00     15.6±0.08ms     2.6 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.00      4.1±0.03ms     4.0 MB/sec    1.01      4.2±0.03ms     4.0 MB/sec
linter/all-rules/numpy/globals.py          1.00    436.2±5.68µs     6.8 MB/sec    1.01    439.7±4.02µs     6.7 MB/sec
linter/all-rules/pydantic/types.py         1.00      6.7±0.02ms     3.8 MB/sec    1.01      6.8±0.02ms     3.8 MB/sec
linter/default-rules/large/dataset.py      1.00      8.1±0.02ms     5.0 MB/sec    1.02      8.3±0.03ms     4.9 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.00   1752.8±8.74µs     9.5 MB/sec    1.02   1796.2±7.75µs     9.3 MB/sec
linter/default-rules/numpy/globals.py      1.00    184.9±1.37µs    16.0 MB/sec    1.01    187.5±2.92µs    15.7 MB/sec
linter/default-rules/pydantic/types.py     1.00      3.7±0.02ms     6.9 MB/sec    1.02      3.8±0.01ms     6.7 MB/sec

[MichaReiser]

Cloning the whole stack here (and when adding to dead_scopes is something we should look at at some point.

I think we could reduce the cost of cloning by:

• Having one Vec<ScopeInternal> that stores all existing scopes
• ScopeInternal is a Scope without its id. The is implicit by the position of the Scope in the vector.
• Add a parent: Option<ScopeId> to scope to allow upwards traversal.
• Change ScopeStack to Vec<ScopeId>

I need to look closer into what parents is used for

[charliermarsh]

parents is the stack of statements that are parents of the current statement (so e.g. if we're looking at a function definition inside of a class, it might have a parents of vec![ClassDef { ... }]).

[MichaReiser]

Thx!

[MichaReiser]

I'm surprised that clippy didn't flag this unnecessary clone

[charliermarsh]

Great change.