Fast path for ASCII only identifiers start

[MichaReiser]

Summary

This PR adds a fast path for ASCII only identifier starts.

I noticed during profiling that the binary_search method is called a lot and wondered why. It turned out, it is used by the xid_start function and we call it when lexing identifiers.
However, the function shouldn't be called at all for my test file because it only contains ASCII characters.

This PR "fixes" our is_identifier_start to not call into xid_start if the character is ascii only.

Test Plan

cargo test

I verified that binary_search no longer appears in the profile.

[MichaReiser]

Current dependencies on/for this PR:

• main
  • PR Fast path for ASCII only identifiers start #6609 👈

This comment was auto-generated by Graphite.

[github-actions]

PR Check Results

Ecosystem

✅ ecosystem check detected no changes.

Benchmark

Linux

group                                      main                                   pr
-----                                      ----                                   --
formatter/large/dataset.py                 1.04      3.9±0.02ms    10.6 MB/sec    1.00      3.7±0.02ms    11.0 MB/sec
formatter/numpy/ctypeslib.py               1.01   767.1±16.79µs    21.7 MB/sec    1.00    759.6±2.77µs    21.9 MB/sec
formatter/numpy/globals.py                 1.00     75.1±0.30µs    39.3 MB/sec    1.05     78.6±0.31µs    37.5 MB/sec
formatter/pydantic/types.py                1.01  1516.9±38.48µs    16.8 MB/sec    1.00   1500.6±5.60µs    17.0 MB/sec
linter/all-rules/large/dataset.py          1.00     10.3±0.03ms     4.0 MB/sec    1.01     10.3±0.02ms     3.9 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.00      2.8±0.03ms     6.0 MB/sec    1.01      2.8±0.03ms     6.0 MB/sec
linter/all-rules/numpy/globals.py          1.00    392.3±3.98µs     7.5 MB/sec    1.00    392.0±1.62µs     7.5 MB/sec
linter/all-rules/pydantic/types.py         1.00      5.3±0.01ms     4.8 MB/sec    1.03      5.5±0.10ms     4.7 MB/sec
linter/default-rules/large/dataset.py      1.00      5.5±0.01ms     7.4 MB/sec    1.02      5.5±0.02ms     7.3 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.00   1204.0±4.30µs    13.8 MB/sec    1.02   1231.1±1.48µs    13.5 MB/sec
linter/default-rules/numpy/globals.py      1.00    141.0±1.08µs    20.9 MB/sec    1.00    141.7±0.26µs    20.8 MB/sec
linter/default-rules/pydantic/types.py     1.00      2.5±0.00ms    10.3 MB/sec    1.02      2.5±0.01ms    10.2 MB/sec

Windows

group                                      main                                   pr
-----                                      ----                                   --
formatter/large/dataset.py                 1.02      4.3±0.06ms     9.4 MB/sec    1.00      4.2±0.10ms     9.6 MB/sec
formatter/numpy/ctypeslib.py               1.00   843.4±12.60µs    19.7 MB/sec    1.00   840.1±12.84µs    19.8 MB/sec
formatter/numpy/globals.py                 1.00     85.8±1.92µs    34.4 MB/sec    1.02     87.4±2.42µs    33.8 MB/sec
formatter/pydantic/types.py                1.00  1682.8±22.48µs    15.2 MB/sec    1.01  1702.2±33.51µs    15.0 MB/sec
linter/all-rules/large/dataset.py          1.01     12.8±0.17ms     3.2 MB/sec    1.00     12.8±0.13ms     3.2 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.01      3.5±0.04ms     4.7 MB/sec    1.00      3.5±0.03ms     4.8 MB/sec
linter/all-rules/numpy/globals.py          1.00    440.8±5.96µs     6.7 MB/sec    1.00    440.6±8.59µs     6.7 MB/sec
linter/all-rules/pydantic/types.py         1.00      6.6±0.10ms     3.9 MB/sec    1.00      6.6±0.07ms     3.9 MB/sec
linter/default-rules/large/dataset.py      1.00      7.0±0.06ms     5.8 MB/sec    1.00      7.0±0.06ms     5.8 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.00  1483.8±14.64µs    11.2 MB/sec    1.00  1482.4±17.30µs    11.2 MB/sec
linter/default-rules/numpy/globals.py      1.00    175.5±3.14µs    16.8 MB/sec    1.00    174.8±2.59µs    16.9 MB/sec
linter/default-rules/pydantic/types.py     1.01      3.1±0.05ms     8.1 MB/sec    1.00      3.1±0.03ms     8.2 MB/sec

[Boshen]

Why profile when I have already done it for you ;-)

https://github.com/web-infra-dev/oxc/blob/b346a9ad6debc3016dd288f3c2c9e595b47a0a37/crates/oxc_parser/src/lexer/mod.rs#L345C1-L358

    fn match_char(&mut self, c: char, mut builder: AutoCow<'a>) -> Kind {
        // fast path for single character tokens
        // '{'  '}'  '('  ')'  '['  ']'  ';' ',' ':' '~'
        let size = c as usize;
        if size < 128 {
            let kind = SINGLE_CHAR_TOKENS[size];
            if kind != Kind::Undetermined {
                return kind;
            }
            // fast path for identifiers
            if is_identifier_start_ascii(c) {
                builder.push_matching(c);
                return self.identifier_name_or_keyword(builder);
            }
        }

https://github.com/web-infra-dev/oxc/blob/b346a9ad6debc3016dd288f3c2c9e595b47a0a37/crates/oxc_syntax/src/identifier.rs#L72-L100

// This contains `$` (36) and `_` (95)
pub const ASCII_START: Align64<[bool; 128]> = Align64([
    F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F,
    F, F, F, F, T, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F, F,
    F, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, F, F, F, F, T,
    F, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, T, F, F, F, F, F,
]);

#[inline]
pub fn is_identifier_start_ascii(c: char) -> bool {
    ASCII_START.0[c as usize]
}

/// Section 12.6 Detect `IdentifierStartChar`
#[inline]
pub fn is_identifier_start_all(c: char) -> bool {
    if c.is_ascii() {
        return is_identifier_start_ascii(c);
    }
    is_id_start_unicode(c)
}

[Boshen]

And if you haven't already: replace unicode-xid with https://github.com/dtolnay/unicode-ident

[MichaReiser]

And if you haven't already: replace unicode-xid with dtolnay/unicode-ident

Oh nice, another distraction for the weekend 😅