Add indentation-aware TokenBuilder and Lexer (#1578)

Co-authored-by: Mark Sujew <[email protected]>

packages/langium/src/parser/indentation-aware.ts

@@ -0,0 +1,367 @@
+/******************************************************************************
+ * Copyright 2024 TypeFox GmbH
+ * This program and the accompanying materials are made available under the
+ * terms of the MIT License, which is available in the project root.
+ ******************************************************************************/
+
+import type { CustomPatternMatcherFunc, TokenType, IToken } from 'chevrotain';
+import type { Grammar, TerminalRule } from '../languages/generated/ast.js';
+import type { TokenBuilderOptions } from './token-builder.js';
+import type { LexerResult } from './lexer.js';
+import type { LangiumCoreServices } from '../services.js';
+import { createToken, createTokenInstance, Lexer } from 'chevrotain';
+import { DefaultTokenBuilder } from './token-builder.js';
+import { DefaultLexer, isTokenTypeArray } from './lexer.js';
+
+export interface IndentationTokenBuilderOptions {
+    /**
+     * The name of the token used to denote indentation in the grammar.
+     * A possible definition in the grammar could look like this:
+     * ```langium
+     * terminal INDENT: ':synthetic-indent:';
+     * ```
+     *
+     * @default 'INDENT'
+     */
+    indentTokenName: string;
+    /**
+     * The name of the token used to denote deindentation in the grammar.
+     * A possible definition in the grammar could look like this:
+     * ```langium
+     * terminal DEDENT: ':synthetic-dedent:';
+     * ```
+     *
+     * @default 'DEDENT'
+     */
+    dedentTokenName: string;
+    /**
+     * The name of the token used to denote whitespace other than indentation and newlines in the grammar.
+     * A possible definition in the grammar could look like this:
+     * ```langium
+     * hidden terminal WS: /[ \t]+/;
+     * ```
+     *
+     * @default 'WS'
+     */
+    whitespaceTokenName: string;
+}
+
+export const indentationBuilderDefaultOptions: IndentationTokenBuilderOptions = {
+    indentTokenName: 'INDENT',
+    dedentTokenName: 'DEDENT',
+    whitespaceTokenName: 'WS',
+};
+
+/**
+ * A token builder that is sensitive to indentation in the input text.
+ * It will generate tokens for indentation and dedentation based on the indentation level.
+ *
+ * Inspired by https://github.com/chevrotain/chevrotain/blob/master/examples/lexer/python_indentation/python_indentation.js
+ */
+export class IndentationAwareTokenBuilder extends DefaultTokenBuilder {
+    /**
+     * The stack in which all the previous matched indentation levels are stored
+     * to understand how deep a the next tokens are nested.
+     */
+    protected indentationStack: number[] = [0];
+    readonly options: IndentationTokenBuilderOptions;
+
+    /**
+     * The token type to be used for indentation tokens
+     */
+    readonly indentTokenType: TokenType;
+
+    /**
+     * The token type to be used for dedentation tokens
+     */
+    readonly dedentTokenType: TokenType;
+
+    /**
+     * A regular expression to match a series of tabs and/or spaces.
+     * Override this to customize what the indentation is allowed to consist of.
+     */
+    protected whitespaceRegExp = /[ \t]+/y;
+
+    constructor(options: Partial<IndentationTokenBuilderOptions> = indentationBuilderDefaultOptions) {
+        super();
+        this.options = {
+            ...indentationBuilderDefaultOptions,
+            ...options,
+        };
+
+        this.indentTokenType = createToken({
+            name: this.options.indentTokenName,
+            pattern: this.indentMatcher,
+            line_breaks: false,
+        });
+
+        this.dedentTokenType = createToken({
+            name: this.options.dedentTokenName,
+            pattern: this.dedentMatcher,
+            line_breaks: false,
+        });
+    }
+
+    override buildTokens(grammar: Grammar, options?: TokenBuilderOptions | undefined) {
+        const tokenTypes = super.buildTokens(grammar, options);
+        if (!isTokenTypeArray(tokenTypes)) {
+            throw new Error('Invalid tokens built by default builder');
+        }
+
+        const { indentTokenName, dedentTokenName, whitespaceTokenName } = this.options;
+
+        // Rearrange tokens because whitespace (which is ignored) goes to the beginning by default, consuming indentation as well
+        // Order should be: dedent, indent, spaces
+        let dedent: TokenType | undefined;
+        let indent: TokenType | undefined;
+        let ws: TokenType | undefined;
+        const otherTokens: TokenType[] = [];
+        for (const tokenType of tokenTypes) {
+            if (tokenType.name === dedentTokenName) {
+                dedent = tokenType;
+            } else if (tokenType.name === indentTokenName) {
+                indent = tokenType;
+            } else if (tokenType.name === whitespaceTokenName) {
+                ws = tokenType;
+            } else {
+                otherTokens.push(tokenType);
+            }
+        }
+        if (!dedent || !indent || !ws) {
+            throw new Error('Some indentation/whitespace tokens not found!');
+        }
+        return [dedent, indent, ws, ...otherTokens];
+    }
+
+    /**
+     * Helper function to check if the current position is the start of a new line.
+     *
+     * @param text The full input string.
+     * @param offset The current position at which to check
+     * @returns Whether the current position is the start of a new line
+     */
+    protected isStartOfLine(text: string, offset: number): boolean {
+        return offset === 0 || '\r\n'.includes(text[offset - 1]);
+    }
+
+    /**
+     * A helper function used in matching both indents and dedents.
+     *
+     * @param text The full input string.
+     * @param offset The current position at which to attempt a match
+     * @returns The current and previous indentation levels and the matched whitespace
+     */
+    protected matchWhitespace(text: string, offset: number) {
+        this.whitespaceRegExp.lastIndex = offset;
+        const match = this.whitespaceRegExp.exec(text);
+        return {
+            currIndentLevel: match?.[0].length ?? 0,
+            prevIndentLevel: this.indentationStack.at(-1)!,
+            match,
+        };
+    }
+
+    /**
+     * Helper function to create an instance of an indentation token.
+     *
+     * @param tokenType Indent or dedent token type
+     * @param text Full input string, used to calculate the line number
+     * @param image The original image of the token (tabs or spaces)
+     * @param offset Current position in the input string
+     * @returns The indentation token instance
+     */
+    protected createIndentationTokenInstance(tokenType: TokenType, text: string, image: string, offset: number) {
+        const lineNumber = text.substring(0, offset).split(/\r\n|\r|\n/).length;
+        return createTokenInstance(
+            tokenType,
+            image,
+            offset, offset + image.length,
+            lineNumber, lineNumber,
+            0, image.length,
+        );
+    }
+
+    /**
+     * A custom pattern for matching indents
+     *
+     * @param text The full input string.
+     * @param offset The offset at which to attempt a match
+     * @param tokens Previously scanned Tokens
+     * @param groups Token Groups
+     */
+    protected indentMatcher: CustomPatternMatcherFunc = (text, offset, tokens, _groups) => {
+        const { indentTokenName } = this.options;
+
+        if (!this.isStartOfLine(text, offset)) {
+            return null;
+        }
+
+        const { currIndentLevel, prevIndentLevel, match } = this.matchWhitespace(text, offset);
+
+        if (currIndentLevel <= prevIndentLevel) {
+            // shallower indentation (should be matched by dedent)
+            // or same indentation level (should be matched by whitespace and ignored)
+            return null;
+        }
+
+        this.indentationStack.push(currIndentLevel);
+
+        const indentToken = this.createIndentationTokenInstance(
+            this.indentTokenType,
+            text,
+            match?.[0] ?? indentTokenName,
+            offset,
+        );
+        tokens.push(indentToken);
+
+        // Token already added, let the indentation now be consumed as whitespace and ignored
+        return null;
+    };
+
+    /**
+     * A custom pattern for matching dedents
+     *
+     * @param text The full input string.
+     * @param offset The offset at which to attempt a match
+     * @param tokens Previously scanned Tokens
+     * @param groups Token Groups
+     */
+    protected dedentMatcher: CustomPatternMatcherFunc = (text, offset, tokens, _groups) => {
+        const { dedentTokenName } = this.options;
+
+        if (!this.isStartOfLine(text, offset)) {
+            return null;
+        }
+
+        const { currIndentLevel, prevIndentLevel, match } = this.matchWhitespace(text, offset);
+
+        if (currIndentLevel >= prevIndentLevel) {
+            // bigger indentation (should be matched by indent)
+            // or same indentation level (should be matched by whitespace and ignored)
+            return null;
+        }
+
+        const matchIndentIndex = this.indentationStack.lastIndexOf(currIndentLevel);
+
+        // Any dedent must match some previous indentation level.
+        if (matchIndentIndex === -1) {
+            console.error(`Invalid dedent level ${currIndentLevel} at offset: ${offset}. Current indetation stack: ${this.indentationStack}`);
+            // throwing an error would crash the language server
+            // TODO: find a way to report error diagnostics message
+            return null;
+        }
+
+        const numberOfDedents = this.indentationStack.length - matchIndentIndex - 1;
+
+        for (let i = 0; i < numberOfDedents; i++) {
+            const token = this.createIndentationTokenInstance(
+                this.dedentTokenType,
+                text,
+                match?.[0] ?? dedentTokenName,
+                offset,
+            );
+            tokens.push(token);
+            this.indentationStack.pop();
+        }
+
+        // Token already added, let the dedentation now be consumed as whitespace and ignored
+        return null;
+    };
+
+    protected override buildTerminalToken(terminal: TerminalRule): TokenType {
+        const tokenType = super.buildTerminalToken(terminal);
+        const { indentTokenName, dedentTokenName, whitespaceTokenName } = this.options;
+
+        if (tokenType.name === indentTokenName) {
+            return this.indentTokenType;
+        } else if (tokenType.name === dedentTokenName) {
+            return this.dedentTokenType;
+        } else if (tokenType.name === whitespaceTokenName) {
+            return createToken({
+                name: whitespaceTokenName,
+                pattern: this.whitespaceRegExp,
+                group: Lexer.SKIPPED,
+            });
+        }
+
+        return tokenType;
+    }
+
+    /**
+     * Resets the indentation stack between different runs of the lexer
+     *
+     * @param text Full text that was tokenized
+     * @returns Remaining dedent tokens to match all previous indents at the end of the file
+     */
+    popRemainingDedents(text: string): IToken[] {
+        const remainingDedents: IToken[] = [];
+        while (this.indentationStack.length > 1) {
+            remainingDedents.push(
+                this.createIndentationTokenInstance(this.dedentTokenType, text, this.options.dedentTokenName, text.length)
+            );
+            this.indentationStack.pop();
+        }
+
+        this.indentationStack = [0];
+        return remainingDedents;
+    }
+}
+
+/**
+ * A lexer that is aware of indentation in the input text.
+ * The only purpose of this lexer is to reset the internal state of the {@link IndentationAwareTokenBuilder}
+ * between the tokenization of different text inputs.
+ *
+ * In your module, you can override the default lexer with this one as such:
+ * ```ts
+ * parser: {
+ *    TokenBuilder: () => new IndentationAwareTokenBuilder(),
+ *    Lexer: (services) => new IndentationAwareLexer(services),
+ * }
+ * ```
+ */
+export class IndentationAwareLexer extends DefaultLexer {
+
+    protected readonly indentationTokenBuilder: IndentationAwareTokenBuilder;
+
+    constructor(services: LangiumCoreServices) {
+        super(services);
+        if (services.parser.TokenBuilder instanceof IndentationAwareTokenBuilder) {
+            this.indentationTokenBuilder = services.parser.TokenBuilder;
+        } else {
+            throw new Error('IndentationAwareLexer requires an accompanying IndentationAwareTokenBuilder');
+        }
+    }
+
+    override tokenize(text: string): LexerResult {
+        const result = super.tokenize(text);
+
+        // reset the indent stack between processing of different text inputs
+        const remainingDedents = this.indentationTokenBuilder.popRemainingDedents(text);
+        result.tokens.push(...remainingDedents);
+
+        // remove any "indent-dedent" pair with an empty body as these are typically
+        // added by comments or lines with just whitespace but have no real value
+        const { indentTokenType, dedentTokenType } = this.indentationTokenBuilder;
+        // Use tokenTypeIdx for fast comparison
+        const indentTokenIdx = indentTokenType.tokenTypeIdx;
+        const dedentTokenIdx = dedentTokenType.tokenTypeIdx;
+        const cleanTokens: IToken[] = [];
+        const length = result.tokens.length - 1;
+        for (let i = 0; i < length; i++) {
+            const token = result.tokens[i];
+            const nextToken = result.tokens[i + 1];
+            if (token.tokenTypeIdx === indentTokenIdx && nextToken.tokenTypeIdx === dedentTokenIdx) {
+                i++;
+                continue;
+            }
+
+            cleanTokens.push(token);
+        }
+        // Push last token separately
+        cleanTokens.push(result.tokens[length]);
+        result.tokens = cleanTokens;
+
+        return result;
+    }
+}

packages/langium/src/parser/index.ts

@@ -7,6 +7,7 @@
 export * from './async-parser.js';
 export * from './completion-parser-builder.js';
 export * from './cst-node-builder.js';
+export * from './indentation-aware.js';
 export * from './langium-parser-builder.js';
 export * from './langium-parser.js';
 export * from './lexer.js';