Skip to main content

Phase 7.6. String builtins

FieldValue
MEPMEP-54
StatusLANDED
Started2026-05-28 (GMT+7)
Landed2026-05-28 (GMT+7)
Tracking PR#22573
Commit9e73065a33

Gate​

24 fixtures across three trios: 8 string_methods (str_contains_false, str_contains_true, str_index_simple, str_index_concat, str_index_in_function, str_methods_combined, str_reverse, str_substring), 8 string_extra (str_join_basic, str_join_single, str_lower, str_split_basic, str_split_join, str_split_spaces, str_upper, str_upper_lower), 8 str_convert (str_bool, str_concat, str_float, str_in_condition, str_in_function, str_int, str_list_values, str_string_identity). All 217 transpiler3/go fixtures green.

Lowering decisions​

  • StrIndexExpr -> mochiStrIndex helper (rune-based index; negative bounds wrap to len + i, OOB panics with the runtime panic code, matching the C runtime). Naive s[i] would index UTF-8 bytes and diverge for any non-ASCII string.
  • StrSubstringExpr -> mochiStrSubstring helper (rune-based half-open slice with clamping on both bounds). The clamp catches both s[-1:5] and s[100:200] rather than panicking, matching the C runtime's permissive semantics.
  • StrReverseExpr -> mochiStrReverse helper (reverse by runes so multibyte characters survive the round-trip; reversing by bytes would mangle UTF-8 surrogates).
  • StrSplitExpr -> strings.Split(s, sep). The Go stdlib version's empty-separator behaviour (split into runes) matches Mochi.
  • StrJoinExpr -> strings.Join(xs, sep). Direct delegation; xs is already a []string from the Phase 3.1 list lowerer.

addImport("strings") is registered for split / join. The three mochiStr* helpers are introduced via addHelper and rendered into the emittedHelpers block before main.

Files changed​

FilePurpose
transpiler3/go/lower/expr.goStrIndexExpr, StrSubstringExpr, StrReverseExpr, StrSplitExpr, StrJoinExpr
transpiler3/go/lower/lower.gomochiStrIndex, mochiStrSubstring, mochiStrReverse helper texts
tests/transpiler3/go/fixtures/str_*/24 fixtures

Test set​

  • 24 TestPhase1Hello/str_* subtests.

Closeout notes​

The rune-based pattern was set up in Phase 7.3 (StrLenExpr); Phase 7.6 reuses the rune-conversion trick for every method that touches a character. The cost is a []rune(s) conversion per call (linear in the string length); this is fine for the fixture goldens but a Phase 7.6.1 optimisation pass could memoise the rune slice for repeated method calls on the same string. No fixture motivates it yet.