Files
docmost/packages/base-formula/src/parser.ts
T
Philipinho 4e5bff6d55 feat(ee): bases
Table and kanban UI, formula engine package, and the base-embed editor extension
2026-06-14 01:29:06 +01:00

202 lines
6.5 KiB
TypeScript

import { tokenize, Token, TokenKind } from "./tokenizer";
import { FormulaParseError } from "./error";
import { MAX_PARSE_DEPTH } from "./types";
import type { OpCode } from "./ast";
import type { RawFormulaAST } from "./ast";
/*
* Pratt parser. Top-level entry parses a full expression and then asserts EOF.
* Binary operators are dispatched through a precedence table in `bp` below.
* `prop(...)`, `if(...)`, `and(...)`, `or(...)` are intercepted when an
* identifier is followed by `(` so they become their dedicated AST nodes.
*/
export function parseRaw(src: string): RawFormulaAST {
const tokens = tokenize(src);
const p = new Parser(tokens);
const expr = p.parseExpr(0);
p.expect(TokenKind.EOF, "Expected end of input");
return expr;
}
const BP: Partial<Record<TokenKind, number>> = {
[TokenKind.OR]: 10,
[TokenKind.AND]: 20,
[TokenKind.EQ]: 30, [TokenKind.NEQ]: 30,
[TokenKind.LT]: 40, [TokenKind.GT]: 40,
[TokenKind.LTE]: 40, [TokenKind.GTE]: 40,
[TokenKind.PLUS]: 50, [TokenKind.MINUS]: 50,
[TokenKind.STAR]: 60, [TokenKind.SLASH]: 60, [TokenKind.PERCENT]: 60,
};
const TOK_TO_OP: Partial<Record<TokenKind, OpCode>> = {
[TokenKind.PLUS]: "+", [TokenKind.MINUS]: "-",
[TokenKind.STAR]: "*", [TokenKind.SLASH]: "/", [TokenKind.PERCENT]: "%",
[TokenKind.EQ]: "==", [TokenKind.NEQ]: "!=",
[TokenKind.LT]: "<", [TokenKind.GT]: ">",
[TokenKind.LTE]: "<=", [TokenKind.GTE]: ">=",
};
class Parser {
private i = 0;
private depth = 0;
constructor(private tokens: Token[]) {}
peek(): Token { return this.tokens[this.i]; }
next(): Token { return this.tokens[this.i++]; }
expect(kind: TokenKind, msg: string): Token {
const t = this.peek();
if (t.kind !== kind) {
throw new FormulaParseError([{
code: "UNEXPECTED_TOKEN", message: msg, span: { start: t.start, end: t.end },
}]);
}
return this.next();
}
// Bound recursive descent. Every path that recurses (parens, unary chains,
// binary rhs, call args) funnels through parseExpr/parseUnary, so guarding
// their entry caps the JS stack and turns pathological nesting into a
// catchable FormulaParseError instead of a RangeError.
private enter(): void {
if (++this.depth > MAX_PARSE_DEPTH) {
const t = this.peek();
throw new FormulaParseError([{
code: "DEPTH_EXCEEDED",
message: `Formula nesting too deep (max ${MAX_PARSE_DEPTH})`,
span: { start: t.start, end: t.end },
}]);
}
}
parseExpr(minBp: number): RawFormulaAST {
this.enter();
try {
return this.parseExprInner(minBp);
} finally {
this.depth--;
}
}
private parseExprInner(minBp: number): RawFormulaAST {
let lhs = this.parseUnary();
while (true) {
const tok = this.peek();
if (tok.kind === TokenKind.AND) {
if (BP[TokenKind.AND]! < minBp) break;
this.next();
const rhs = this.parseExpr(BP[TokenKind.AND]! + 1);
lhs = { t: "and", args: [lhs, rhs] };
continue;
}
if (tok.kind === TokenKind.OR) {
if (BP[TokenKind.OR]! < minBp) break;
this.next();
const rhs = this.parseExpr(BP[TokenKind.OR]! + 1);
lhs = { t: "or", args: [lhs, rhs] };
continue;
}
const bp = BP[tok.kind];
if (bp == null || bp < minBp) break;
this.next();
const rhs = this.parseExpr(bp + 1);
const op = TOK_TO_OP[tok.kind]!;
lhs = { t: "op", op, args: [lhs, rhs] };
}
return lhs;
}
parseUnary(): RawFormulaAST {
const tok = this.peek();
if (tok.kind === TokenKind.MINUS) {
this.next();
this.enter();
try {
return { t: "op", op: "neg", args: [this.parseUnary()] };
} finally {
this.depth--;
}
}
if (tok.kind === TokenKind.NOT) {
this.next();
this.enter();
try {
return { t: "op", op: "not", args: [this.parseUnary()] };
} finally {
this.depth--;
}
}
return this.parsePrimary();
}
parsePrimary(): RawFormulaAST {
const tok = this.next();
switch (tok.kind) {
case TokenKind.NUMBER: return { t: "num", v: Number(tok.text) };
case TokenKind.STRING: return { t: "str", v: tok.text };
case TokenKind.TRUE: return { t: "bool", v: true };
case TokenKind.FALSE: return { t: "bool", v: false };
case TokenKind.NULL: return { t: "null" };
case TokenKind.LPAREN: {
const e = this.parseExpr(0);
this.expect(TokenKind.RPAREN, "Expected ')'");
return e;
}
case TokenKind.AND:
case TokenKind.OR:
case TokenKind.IDENT: {
if (this.peek().kind !== TokenKind.LPAREN) {
throw new FormulaParseError([{
code: "UNEXPECTED_TOKEN",
message: `Unexpected identifier '${tok.text}' (did you mean prop("${tok.text}")?)`,
span: { start: tok.start, end: tok.end },
}]);
}
this.next(); // LPAREN
const args: RawFormulaAST[] = [];
if (this.peek().kind !== TokenKind.RPAREN) {
args.push(this.parseExpr(0));
while (this.peek().kind === TokenKind.COMMA) {
this.next();
args.push(this.parseExpr(0));
}
}
this.expect(TokenKind.RPAREN, "Expected ')'");
// Match special-form/keyword names case-insensitively; the `call`
// node below keeps the raw casing the user typed.
const head = tok.text.toLowerCase();
if (head === "prop") {
if (args.length !== 1 || args[0].t !== "str") {
throw new FormulaParseError([{
code: "UNEXPECTED_TOKEN",
message: 'prop() expects exactly one string literal argument',
span: { start: tok.start, end: tok.end },
}]);
}
return { t: "propName", name: args[0].v };
}
if (head === "if") {
if (args.length !== 3) {
throw new FormulaParseError([{
code: "ARITY_MISMATCH",
message: "if() expects exactly 3 arguments",
span: { start: tok.start, end: tok.end },
}]);
}
return { t: "if", cond: args[0], then: args[1], else: args[2] };
}
if (head === "and") return { t: "and", args };
if (head === "or") return { t: "or", args };
return { t: "call", fn: tok.text, args };
}
default:
throw new FormulaParseError([{
code: "UNEXPECTED_TOKEN",
message: `Unexpected token '${tok.text || tok.kind}'`,
span: { start: tok.start, end: tok.end },
}]);
}
}
}