Theia-Modula-2/grammars/Oberon2/README.md

Oberon-2 validator pilot

oberon2_oberon.atg describes Oberon-2 (Wirth/Mössenböck). It generated cleanly with Coco/R V1.53 (CR) but needed three repairs before GNU Modula-2 (gm2 -fiso) could build it and real programs would parse — all diagnosed against voc as oracle:

1. Duplicate CASE labels in the generated statement procedure: ident-led assignment vs ProcedureCall merged into a single IdentTail dispatch (":=" vs "(", plus selectors).
2. The classic optional-prefix trap in qualident: [ident "."] committed on any identifier and then demanded ., so every TYPE T = INTEGER failed (same fix as the Oberon-07 pilot: ident { "." ident }).
3. The reference grammar has no FOR statement at all — added FOR ident := ... TO ... [BY ...] DO ... END (verified against voc, which accepts BY -1).
4. The reference string token used " quotes; Oberon strings use ' — fixed token and noQuote set.
5. The designator's type-guard alternative "(" qualident ")" shadowed calls with non-identifier arguments (f(1) failed while f(i) parsed as a guard); merged to "(" [ ExpList ] ")".
6. Type-bound procedures gained their receiver: PROCEDURE [(VAR) v: T] [Name*] ... (verified against voc).

Two benign LL(1) warning families remain (dangling-ELSE-style, loop-exit greediness — both resolve correctly).

Oberon-2 facts confirmed by probe (validator matches voc): no enumerations, WITH only in guarded-pointer form, CASE with | separators and ELSE, LOOP/EXIT, RETURN with optional value, &/~/OR, IN/IS, sets with .. ranges, NIL, NEW (no dispose), DIV/MOD, IMPORT lists, FORWARD (PROCEDURE ^ name) declarations.

Note: the reference Grammars/Voc/voc_oberon.atg is byte-identical to the Oberon-2 grammar (modulo its header comment), and Grammars/Oberon-grammar/original_oberon.atg differs only in its tags line — this single validator covers all three.

Build

./build.sh        # -> build/O2 (build/ is git-ignored)

Needs CR (Coco/R), gm2 -fiso, the parser.frm/scanner.frm frames and FileIO.def/.mod — all resolved from the environment with sensible defaults (CRFRAMES, M2LIB, CR, GM2).

Differential results (O2 vs voc)

Accept battery (22 cases: all statements, FOR/BY, CASE/ELSE, LOOP/EXIT, guarded WITH, NEW/deref, ORD, floats, strings, Out calls, RETURN) + tests/deep.Mod (records, pointers, multi-dim arrays, nested routines, type-bound procedures, all statements): 0 false positives. Broken mutants (9): 0 missed, positions close. (i := 1 without ; is accepted by voc too — separator leniency, correctly skipped.)