Theia-Modula-2/grammars/P4

P4 Pascal validator pilot

p4_pascal.atg describes Wirth's Pascal-P4 (1977) compiler input. It needed a full repair pass to build with Coco/R V1.53 (CR) and GNU Modula-2 (gm2 -fiso), all diagnosed against fpc -Miso as oracle:

1. Bare ; is not a Coco/R literal — every statement separator is written ";" (this alone was 5 of the original errors; the same applies inside (...) groups).
2. The Statement production was missing its terminating ..
3. EmptyStatement = "" is illegal — empty positions use the [ X { ";" [ X ] } ] list shape instead (progress-guaranteed, so the generated parser cannot hang on empty matches).
4. ident cannot be both a TOKENS class and a production (renamed the literal tokens to pnumber/pstring); EOF is reserved (dropped the junk line with it); the start production must match the COMPILER name.
5. Left-recursive Primary "^" rewritten as a postfix loop.
6. Undefined nonterminals filled in: element productions extracted from the ...List wrappers (ConstDecl, TypeDecl, VarDecl), ProcDecl (nested procedures/functions), Value, PointerType, noQuote1, ParList→ParamList, Stmts→Statements, CaseStatement/WhileStatement/...→actual names, Var→VarIdent.
7. Name collisions in generated CASE statements fixed by merging: duplicate for..to/downto branches, duplicate Type/StructType entries (ArrayType/SetType/ProcType reachable twice), and ident-led Assignment vs ProcCall (single IdentTail dispatch on := vs ().
8. Grammar gaps closed to oracle level: TypeIdent (named types — without it every VAR x: MyType failed), Compound program bodies, trailing-;-tolerant sections/fields/case-items, {} comments, subscripts/field selectors in statements and expressions, and/or words alongside P4 &, bare parameterless calls, statement labels + goto (+LabelSection), and-less or..., formal parameter groups (name: type, var-prefixed) for declarations (call sites keep expression lists), subrange array indices (1..10), const expressions as Value, CASE branches take a single statement (also fixes a ;-separator competition), multi-routine { ProcDecl } blocks.

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

Floats (1.5E-3), enums, subranges (0..255, N..M, dotted names) and Label = pnumber | ident (enum case labels; statement labels stay numeric-only or i := 1 breaks on the :=) were added to oracle level afterwards.

if/while/for branches take a single Statement (standard Pascal — plural branches greedily swallow the ; separating case items, so if..then begin..end; inside a case branch broke), and begin..end compounds are statements in their own right.

The generated parser gets a strict EOF check (IF sym # 0 THEN SynError(0) after the start symbol in Parse, patched in by build.sh with a loud assertion): without it, the Coco/R driver never scans past the program's final ., silently ignoring trailing garbage — and worse, masking any failure that leaves a . behind. This makes the validator marginally stricter than fpc, which accepts trailing junk after END. — degenerate input only, documented here.

Build

./build.sh        # -> build/P4 (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 (P4 vs fpc -Miso)

Accept battery + tests/deep.pas (labels/goto, nested routines with var params, records, enums, subranges, sets, pointers, all statements, hex-free literals): 0 false positives. Broken mutants: 0 missed, positions close. P4-era shapes (&, #, unit sections, return) are accepted leniently although fpc -Miso rejects them (direction is accept-more, so these can only ever be missed rejections, never false positives).