bytes/internal/regex_engine/ and string/internal/regex_engine/ contain byte-identical copies of: symbolize.mbt (42 lines), compile.mbt (43), using.mbt (49), ast/pattern.mbt (205), ast/using.mbt (13), symbol_map/using.mbt (13) — roughly 365 lines duplicated verbatim that must be kept in sync by hand (translate.mbt and the .mbti files legitimately differ, and drift has already crept in: the string/ copy of translate.mbt used an imperative loop where the bytes/ twin uses a comprehension — realigned in #3821).
Both packages already depend on the shared internal/regex_engine (automata/ + shared_types/), so the identical files look liftable into a shared internal package. This is an architectural consolidation rather than a mechanical dedup — the byte/string engines are parameterized differently at the edges — so filing for design discussion rather than PRing directly.
bytes/internal/regex_engine/andstring/internal/regex_engine/contain byte-identical copies of:symbolize.mbt(42 lines),compile.mbt(43),using.mbt(49),ast/pattern.mbt(205),ast/using.mbt(13),symbol_map/using.mbt(13) — roughly 365 lines duplicated verbatim that must be kept in sync by hand (translate.mbtand the.mbtifiles legitimately differ, and drift has already crept in: thestring/copy oftranslate.mbtused an imperative loop where thebytes/twin uses a comprehension — realigned in #3821).Both packages already depend on the shared
internal/regex_engine(automata/+shared_types/), so the identical files look liftable into a shared internal package. This is an architectural consolidation rather than a mechanical dedup — the byte/string engines are parameterized differently at the edges — so filing for design discussion rather than PRing directly.