pli-f — a PL/I F parser that refuses to be told what words mean

PL/I has no reserved words. This parser made peace with that. Mostly.

A clean-room parser for PL/I F, the 1965 IBM dialect described in IBM Operating System/360 PL/I: Language Specifications (form C28-6571). Written in OCaml with menhir. It started life as a parser for XPL and just kept going until it spoke full PL/I F: structures, PICTURE, pointers, BASED storage, FILE I/O, ON-units, the lot.

The party trick (Ta daaah!)

KEYWORDS: PROCEDURE;
   DECLARE FORMAT  FIXED;
   DECLARE RECORD  FIXED;
   DECLARE DISPLAY FIXED;
   FORMAT  = 1;
   RECORD  = 2;
   DISPLAY = FORMAT + RECORD;   /* yes, those are all keywords. yes, it parses. */
   RETURN;
END KEYWORDS;

PL/I reserves nothing. You can name a variable FORMAT, assign it to RECORD, and pass it to a procedure called ENTRY, and the language just lets you. The 1965 committee looked at the concept of reserved keywords and said "no thanks."

So the parser does a little dance: the lexer eagerly turns every keyword into a token, and then the grammar laboriously turns a big chunk of them back into identifiers, by context, wherever a name is allowed. The lexer giveth, the grammar taketh away. (Structural words like IF/THEN/DO/END are the one exception — let those double as variable names and the grammar catches fire.)

Badge of honour

Building this grammar, menhir reports 654 shift/reduce conflicts, arbitrarily resolved. That is not sloppiness. That is the bill for "every keyword is also a valid variable name," itemised. A grammar that can't tell a keyword from a name until it sees what comes next is conflict-heavy by construction, and menhir just picks the sensible branch a few hundred times. It's bloody good!

What it parses

• Procedures — params, return types, RECURSIVE, nested, END label
• DECLARE — FIXED / FLOAT / CHARACTER / BIT / POINTER, precision, arrays with bounds, INITIAL, LITERALLY, identifier lists, LIKE, DEFINED
• Structures — level numbers (1 EMPLOYEE, 2 NAME ...)
• PICTURE specifications
• Storage — STATIC / AUTOMATIC / CONTROLLED, BASED, ALLOCATE / FREE
• Control flow — IF/THEN/ELSE, DO WHILE/UNTIL/CASE/step TO...BY, SELECT/WHEN/OTHERWISE, GO TO, labels
• Expressions — arithmetic, **, concatenation ||, ^ not, relations, subscripts, A.B member access
• I/O — GET / PUT / READ / WRITE / OPEN / FILE / FORMAT
• Conditions — ON / REVERT / SIGNAL
• And the headline feature: none of those names are reserved

Build and run

dune build
dune test                                         # the unit suite
dune exec bin/main.exe -- test/corpus/*.pli        # parse files, report a count
dune exec bin/main.exe -- --dump path/to/prog.pli  # pretty-print the AST

test/corpus/ has small programs to poke at, including 04_no_reserved_words.pli if you want to watch the party trick run.

About the spec

The grammar follows IBM's PL/I: Language Specifications, form C28-6571 (1965). That manual is IBM-copyrighted Systems Reference Library material, so it is not included in this repo. If you want to read along, the scanned PDF lives on bitsavers.

Heritage

This grew out of a parser for XPL, the little PL/I-subset language compilers were once written in. It cut its teeth on real XPL source before being pushed up to full PL/I F. (Some XPL-isms survive as accepted synonyms, e.g. FUNCTION for PROCEDURE, CLOSE for END.)

Sitrep

It parses, and it parses a genuinely large slice of the language. It is a parser — front end only, AST out, no semantic analysis or codegen. If you feed it valid PL/I F it builds you a tree; that is the whole job and it does it. It's how I made my HALMAT project work, so if you have any legacy NASA code, there's a chance it could work, as a substantial portion of it was originally built on a subset of PL/I.

Licence

Apache 2.0.