nandi.uk
Fork of daniellemaywood.uk/gleam — Wasm codegen work
Optimise single character matching on JavaScript
author
Gears
committer
Louis Pilfold
date
(Mar 26, 2026, 12:29 PM UTC)
commit
489081e1
489081e179c7dd3564b09a4b1143c30c998af34b
parent
27309fb4
27309fb4e9521a24d1598773f0d38225b9d98c0c
change-id
llzlrsvt
llzlrsvtqttwslqopuyxuqsroroookmw
+223
-13
+16
CHANGELOG.md
+16
CHANGELOG.md
···
49
49
50
50
([Giacomo Cavalieri](https://github.com/giacomocavalieri))
51
51
52
+
- The compiler now emits more efficient code when matching on single-character
53
+
string prefixes on the JavaScript target. For example, the `glance` package
54
+
is now nearly 30% faster on the JavaScript target:
55
+
56
+
```
57
+
# before:
58
+
min: 10.8ms, max: 365.82ms, median: 14.74ms, mean: 14.76ms
59
+
warmup: 100/1.5s, total post-warmup: 1000/14.76s
60
+
61
+
# after:
62
+
min: 8.96ms, max: 143.76ms, median: 10.72ms, mean: 11.06ms
63
+
warmup: 100/1.24s, total post-warmup: 1000/11.06s
64
+
```
65
+
66
+
([Surya Rose](https://github.com/GearsDatapacks))
67
+
52
68
### Build tool
53
69
54
70
- The `gleam hex owner add` command has been added, which allows adding
+78
-1
compiler-core/src/javascript/decision.rs
+78
-1
compiler-core/src/javascript/decision.rs
···
408
408
assignments.push(let_doc(name, value.to_doc()))
409
409
};
410
410
411
+
// Variable storing the character code for the first character of a string.
412
+
// This is only declared if multiple patterns match on just the first
413
+
// character, as it allows us to avoid calling `.startsWith` multiple times
414
+
// and just call `.charCodeAt(0)` once, which is much faster.
415
+
let first_character_variable = if multiple_single_character_prefix_checks(choices) {
416
+
let name = self.variables.next_local_var(&ASSIGNMENT_VAR.into());
417
+
let string = self.variables.get_value(var);
418
+
let first_character = docvec![string, ".charCodeAt(0)"];
419
+
assignments.push(let_doc(name.clone(), first_character));
420
+
Some(name)
421
+
} else {
422
+
None
423
+
};
424
+
411
425
let mut if_ = CaseBody::Statements(nil());
412
426
for (i, (check, decision)) in choices.iter().enumerate() {
413
427
self.variables.record_check_assignments(var, check);
···
419
433
// referenced by this check
420
434
let (check_doc, body, mut segment_assignments) = self.inside_new_scope(|this| {
421
435
let segment_assignments = this.variables.bit_array_segment_assignments(check);
422
-
let check_doc = this.variables.runtime_check(var, check);
436
+
437
+
// If the pattern matches on a single character, use the character
438
+
// code instead of `.startsWith`.
439
+
let check_doc = if let Some(code) = single_character_prefix_code(check) {
440
+
let first_character = if let Some(variable) = &first_character_variable {
441
+
variable.to_doc()
442
+
} else {
443
+
// This is the only single-character match in this `case`
444
+
// expression, so we don't bind it to a variable and just
445
+
// call `.charCodeAt` inline. This is still faster than
446
+
// `.startsWith`.
447
+
let string = this.variables.get_value(var);
448
+
docvec![string, ".charCodeAt(0)"]
449
+
};
450
+
451
+
docvec![first_character, " === ", code]
452
+
} else {
453
+
this.variables.runtime_check(var, check)
454
+
};
455
+
423
456
let body = this.decision(decision);
424
457
(check_doc, body, segment_assignments)
425
458
});
···
700
733
],
701
734
)
702
735
}
736
+
}
737
+
738
+
/// Returns the character code for the character being matched for patterns matching
739
+
/// on single-character string prefixes.
740
+
fn single_character_prefix_code(check: &RuntimeCheck) -> Option<u32> {
741
+
match check {
742
+
// On JavaScript, a single "character" is one that can be represented as
743
+
// a single UTF-16 codepoint.
744
+
RuntimeCheck::StringPrefix { prefix, rest } if utf16_no_escape_len(prefix) == 1 => {
745
+
convert_string_escape_chars(prefix)
746
+
.chars()
747
+
.next()
748
+
.map(|first| first as u32)
749
+
}
750
+
RuntimeCheck::Int { .. }
751
+
| RuntimeCheck::Float { .. }
752
+
| RuntimeCheck::String { .. }
753
+
| RuntimeCheck::StringPrefix { .. }
754
+
| RuntimeCheck::Tuple { .. }
755
+
| RuntimeCheck::BitArray { .. }
756
+
| RuntimeCheck::Variant { .. }
757
+
| RuntimeCheck::NonEmptyList { .. }
758
+
| RuntimeCheck::EmptyList => None,
759
+
}
760
+
}
761
+
762
+
/// Returns whether a `case` expression contains multiple patterns matching on
763
+
/// the first character of a string.
764
+
fn multiple_single_character_prefix_checks(choices: &[(RuntimeCheck, Decision)]) -> bool {
765
+
let mut encountered_check = false;
766
+
767
+
for (check, _) in choices.iter() {
768
+
if let RuntimeCheck::StringPrefix { prefix, .. } = check
769
+
&& utf16_no_escape_len(prefix) == 1
770
+
{
771
+
if encountered_check {
772
+
return true;
773
+
} else {
774
+
encountered_check = true;
775
+
}
776
+
}
777
+
}
778
+
779
+
false
703
780
}
704
781
705
782
pub fn let_<'a>(
+25
compiler-core/src/javascript/tests/case.rs
+25
compiler-core/src/javascript/tests/case.rs
···
943
943
}"#
944
944
)
945
945
}
946
+
947
+
// https://github.com/gleam-lang/gleam/issues/5400
948
+
#[test]
949
+
fn case_matching_single_character_prefixes() {
950
+
assert_js!(
951
+
r#"
952
+
pub fn parse_digit(s: String) -> Int {
953
+
case s {
954
+
"0" <> s -> 0
955
+
"1" <> s -> 1
956
+
"2" <> s -> 2
957
+
"3" <> s -> 3
958
+
"4" <> s -> 4
959
+
"5" <> s -> 5
960
+
"6" <> s -> 6
961
+
"7" <> s -> 7
962
+
"8" <> s -> 8
963
+
"9" <> s -> 9
964
+
"" -> todo
965
+
_ -> panic
966
+
}
967
+
}
968
+
"#
969
+
);
970
+
}
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__assignments__let_assert_nested_string_prefix.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__assignments__let_assert_nested_string_prefix.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__case_building_matched_string_1.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__case_building_matched_string_1.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__case_building_matched_string_2.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__case_building_matched_string_2.snap
+83
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__case_matching_single_character_prefixes.snap
+83
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__case_matching_single_character_prefixes.snap
···
1
+
---
2
+
source: compiler-core/src/javascript/tests/case.rs
3
+
expression: "\npub fn parse_digit(s: String) -> Int {\n case s {\n \"0\" <> s -> 0\n \"1\" <> s -> 1\n \"2\" <> s -> 2\n \"3\" <> s -> 3\n \"4\" <> s -> 4\n \"5\" <> s -> 5\n \"6\" <> s -> 6\n \"7\" <> s -> 7\n \"8\" <> s -> 8\n \"9\" <> s -> 9\n \"\" -> todo\n _ -> panic\n }\n}\n"
4
+
---
5
+
----- SOURCE CODE
6
+
7
+
pub fn parse_digit(s: String) -> Int {
8
+
case s {
9
+
"0" <> s -> 0
10
+
"1" <> s -> 1
11
+
"2" <> s -> 2
12
+
"3" <> s -> 3
13
+
"4" <> s -> 4
14
+
"5" <> s -> 5
15
+
"6" <> s -> 6
16
+
"7" <> s -> 7
17
+
"8" <> s -> 8
18
+
"9" <> s -> 9
19
+
"" -> todo
20
+
_ -> panic
21
+
}
22
+
}
23
+
24
+
25
+
----- COMPILED JAVASCRIPT
26
+
import { makeError } from "../gleam.mjs";
27
+
28
+
const FILEPATH = "src/module.gleam";
29
+
30
+
export function parse_digit(s) {
31
+
let $ = s.charCodeAt(0);
32
+
if ($ === 48) {
33
+
let s$1 = s.slice(1);
34
+
return 0;
35
+
} else if ($ === 49) {
36
+
let s$1 = s.slice(1);
37
+
return 1;
38
+
} else if ($ === 50) {
39
+
let s$1 = s.slice(1);
40
+
return 2;
41
+
} else if ($ === 51) {
42
+
let s$1 = s.slice(1);
43
+
return 3;
44
+
} else if ($ === 52) {
45
+
let s$1 = s.slice(1);
46
+
return 4;
47
+
} else if ($ === 53) {
48
+
let s$1 = s.slice(1);
49
+
return 5;
50
+
} else if ($ === 54) {
51
+
let s$1 = s.slice(1);
52
+
return 6;
53
+
} else if ($ === 55) {
54
+
let s$1 = s.slice(1);
55
+
return 7;
56
+
} else if ($ === 56) {
57
+
let s$1 = s.slice(1);
58
+
return 8;
59
+
} else if ($ === 57) {
60
+
let s$1 = s.slice(1);
61
+
return 9;
62
+
} else if (s === "") {
63
+
throw makeError(
64
+
"todo",
65
+
FILEPATH,
66
+
"my/mod",
67
+
14,
68
+
"parse_digit",
69
+
"`todo` expression evaluated. This code has not yet been implemented.",
70
+
{}
71
+
)
72
+
} else {
73
+
throw makeError(
74
+
"panic",
75
+
FILEPATH,
76
+
"my/mod",
77
+
15,
78
+
"parse_digit",
79
+
"`panic` expression evaluated.",
80
+
{}
81
+
)
82
+
}
83
+
}
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__deeply_nested_string_prefix_assignment.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__deeply_nested_string_prefix_assignment.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__nested_string_prefix_assignment.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__nested_string_prefix_assignment.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__slicing_is_handled_properly_with_multiple_branches.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__case__slicing_is_handled_properly_with_multiple_branches.snap
-2
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__strings__case.snap
-2
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__strings__case.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__strings__string_prefix_assignment_with_multiple_subjects.snap
+1
-1
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__strings__string_prefix_assignment_with_multiple_subjects.snap
+3
-3
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__strings__string_prefix_utf16.snap
+3
-3
compiler-core/src/javascript/tests/snapshots/gleam_core__javascript__tests__strings__string_prefix_utf16.snap
···
31
31
----- COMPILED JAVASCRIPT
32
32
export function go(x) {
33
33
let $ = "Θ wibble wobble";
34
-
if ($.startsWith("Θ")) {
34
+
if ($.charCodeAt(0) === 920) {
35
35
let rest = $.slice(1);
36
36
rest
37
37
} else {
···
52
52
""
53
53
}
54
54
let $3 = "\" is a an escaped quote";
55
-
if ($3.startsWith("\"")) {
55
+
if ($3.charCodeAt(0) === 34) {
56
56
let rest = $3.slice(1);
57
57
rest
58
58
} else {
59
59
""
60
60
}
61
61
let $4 = "\\ is a an escaped backslash";
62
-
if ($4.startsWith("\\")) {
62
+
if ($4.charCodeAt(0) === 92) {
63
63
let rest = $4.slice(1);
64
64
return rest;
65
65
} else {
+11
test/language/test/language/string_pattern_matching_test.gleam
+11
test/language/test/language/string_pattern_matching_test.gleam
···
61
61
}
62
62
assert " is a newline that escaped" == string_2
63
63
}
64
+
65
+
pub fn multiple_single_character_matches() {
66
+
let string = "1234"
67
+
let string_2 = case string {
68
+
"0" <> _ -> panic
69
+
"1" <> rest -> rest
70
+
"2" <> _ -> panic
71
+
_ -> panic
72
+
}
73
+
assert "234" == string_2
74
+
}