Theia-Modula-2/grammars/PascalS

Pascal-S validator pilot

pascals.atg describes Wirth Pascal-S (1976) as implemented by PascalS0 (Turbo-Pascal batch interpreter). It was authored from the P2 pilot skeleton and corrected against some 90 oracle probes — every construct below was checked against the running interpreter (no fpc involved anywhere).

Oracle notes

psc was built with fpc -Mtp from Pascals/ plus a /tmp-only TrmInOut stub (the shipped unit needs DOS REGISTERS/MsDos; only console setup, irrelevant to parse verdicts). Behaviors the harness respects, all verified:

• Verdicts come from the .lst listing: ******* ^ NNN markers (full error table printed in PascalS0.pas:errormsg); absence = accept. Feeding is via psc src.lst args with stdin </dev/null (renders blocking on read harmless); 15s timeout (valid programs finish instantly).
• Lowercase is accepted (Turbo scanner), so the validator uses IGNORE CASE — unlike the P2 pilot.
• Mandatory program heading with file list: program t (input, output); (bare program t; is error 9). input/output exist for eof/eoln only.

Key probed Pascal-S facts: and/or/not/div/mod words; if/while/repeat/for to/downto/case (no ELSE, no OTHERWISE); (* *) comments that do not nest, plus {}; const/type/var/nested routines with var params; integer, real, char, boolean (no string/enum/subrange/pointer/set/file types); multi-dim arrays with const bounds; records without variants; read/readln/write/writeln (bare writeln; is legal!); write-widths; odd/abs/ord/chr; eof/eoln with mandatory file arg; unary signs only at expression start (a + +b is error 58, but -a is fine); NOT anywhere; single-statement branches; empty statements; char/enum-less case labels (pnumber/ident/pstring).

Absent (all probed): labels/goto, forward, packed, sets, pointers (^ is error 24), files, with, procedural params, break/exit/halt/return, nil, _ in identifiers (error 24), // comments, OTHERWISE, max, const expressions (3 + 4 is error 6), multi-char const strings (error 50), 1./.5 reals, $/# literals, IN, chained assignment, function-calls-as-statements (error 45).

Grammar notes

Single-statement branches (avoids ;-capture), begin..end as statements, trailing-;-tolerant sections/fields/case-items, order-free declaration sections, no label section at all, EofCall with mandatory (file) (which is what rejects bare eof), strict EOF check (patched into Parse by build.sh, which also fails on degenerate IF/WHILE FALSE generation).

Lenient acceptances (documented, never false positives): multi-char const strings (a pchar token breaks V1.53 determinism — probe H2 shows the oracle's error 50); bare builtin calls (odd; parses although only eof/eoln strictness was probed); TYPE-level constructs the oracle would mark semantically are out of scope for a syntax validator.

Build

./build.sh        # -> build/PascalS (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 (PascalS vs psc)

Accept battery (22 cases: all statements, arrays, records, ord/chr, reals, strings, odd, write-widths, eof(input)): 0 false positives. Broken mutants (12: truncation, bad factors, unterminated comment, @, missing TO, ELSE, goto, _, 1..3, set constructor): 0 missed, positions close. Example corpus (Pascals/eg0–eg7): agree on all 8 (eg0/eg1 are intentionally broken and both sides reject them). tests/deep.pas (nested routines with var params, records, all statements, reals): clean on both sides.