All Downloads are FREE. Search and download functionalities are using the official Maven repository.

com.google.javascript.jscomp.CodeGenerator Maven / Gradle / Ivy

There is a newer version: 9.0.8
Show newest version
/*
 * Copyright 2004 The Closure Compiler Authors.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package com.google.javascript.jscomp;

import static com.google.common.base.Preconditions.checkNotNull;
import static com.google.common.base.Preconditions.checkState;

import com.google.common.base.Ascii;
import com.google.common.base.Preconditions;
import com.google.debugging.sourcemap.Util;
import com.google.javascript.jscomp.parsing.parser.FeatureSet;
import com.google.javascript.jscomp.parsing.parser.FeatureSet.Feature;
import com.google.javascript.jscomp.parsing.parser.util.SourcePosition;
import com.google.javascript.rhino.JSDocInfo;
import com.google.javascript.rhino.JSDocInfo.Visibility;
import com.google.javascript.rhino.Node;
import com.google.javascript.rhino.NonJSDocComment;
import com.google.javascript.rhino.QualifiedName;
import com.google.javascript.rhino.Token;
import com.google.javascript.rhino.TokenStream;
import javax.annotation.Nullable;

/** CodeGenerator generates codes from a parse tree, sending it to the specified CodeConsumer. */
public class CodeGenerator {
  private static final String LT_ESCAPED = "\\x3c";
  private static final String GT_ESCAPED = "\\x3e";

  private final CodeConsumer cc;

  private final OutputCharsetEncoder outputCharsetEncoder;

  private final boolean preferSingleQuotes;
  private final boolean preserveTypeAnnotations;
  private final boolean printNonJSDocComments;
  private final boolean trustedStrings;
  private final boolean quoteKeywordProperties;
  private final boolean useOriginalName;
  private final FeatureSet outputFeatureSet;
  private final JSDocInfoPrinter jsDocInfoPrinter;

  private CodeGenerator(CodeConsumer consumer) {
    cc = consumer;
    outputCharsetEncoder = null;
    preferSingleQuotes = false;
    trustedStrings = true;
    preserveTypeAnnotations = false;
    printNonJSDocComments = false;
    quoteKeywordProperties = false;
    useOriginalName = false;
    this.outputFeatureSet = FeatureSet.BARE_MINIMUM;
    this.jsDocInfoPrinter = new JSDocInfoPrinter(false);
  }

  protected CodeGenerator(CodeConsumer consumer, CompilerOptions options) {
    cc = consumer;

    this.outputCharsetEncoder = new OutputCharsetEncoder(options.getOutputCharset());
    this.preferSingleQuotes = options.preferSingleQuotes;
    this.trustedStrings = options.trustedStrings;
    this.preserveTypeAnnotations = options.preserveTypeAnnotations;
    this.printNonJSDocComments = options.getPreserveNonJSDocComments();
    this.quoteKeywordProperties = options.shouldQuoteKeywordProperties();
    this.useOriginalName = options.getUseOriginalNamesInOutput();
    this.outputFeatureSet = options.getOutputFeatureSet();
    this.jsDocInfoPrinter = new JSDocInfoPrinter(useOriginalName);
  }

  static CodeGenerator forCostEstimation(CodeConsumer consumer) {
    return new CodeGenerator(consumer);
  }

  /** Insert a top-level identifying file as .i.js generated typing file. */
  void tagAsTypeSummary() {
    add("/** @fileoverview @typeSummary */\n");
  }

  /** Insert a ECMASCRIPT 5 strict annotation. */
  public void tagAsStrict() {
    add("'use strict';");
    cc.endLine();
  }

  private void printJSDocComment(Node node, JSDocInfo jsDocInfo) {
    String jsdocAsString = jsDocInfoPrinter.print(jsDocInfo);
    // Don't print an empty jsdoc
    if (!jsdocAsString.equals("/** */ ")) {
      add(jsdocAsString);
      if (!node.isCast()) {
        cc.endLine();
      }
    }
  }

  private void printNonJSDocComment(Node node, NonJSDocComment nonJSDocComment) {
    String nonJSDocCommentString = nonJSDocComment.getCommentString();
    if (!nonJSDocCommentString.isEmpty()) {
      addNonJsDoc_nonTrailing(node, nonJSDocComment);
    }
  }

  /**
   * Print Leading JSDocComments or NonJSDocComments for the given node in order, depending on their
   * source location.
   */
  private void printLeadingCommentsInOrder(Node node) {
    JSDocInfo jsDocInfo = node.getJSDocInfo();
    NonJSDocComment nonJSDocComment = node.getNonJSDocComment();

    boolean printJSDoc = preserveTypeAnnotations && jsDocInfo != null;
    boolean printNonJSDoc = printNonJSDocComments && nonJSDocComment != null;
    if (printJSDoc && printNonJSDoc) {
      if (jsDocInfo.getOriginalCommentPosition() < nonJSDocComment.getStartPosition().getOffset()) {
        printJSDocComment(node, jsDocInfo);
        printNonJSDocComment(node, nonJSDocComment);
      } else {
        printNonJSDocComment(node, nonJSDocComment);
        printJSDocComment(node, jsDocInfo);
      }
      return;
    }

    if (printJSDoc) {
      printJSDocComment(node, jsDocInfo);
      return;
    }

    if (printNonJSDoc) {
      printNonJSDocComment(node, nonJSDocComment);
      return;
    }
  }

  /**
   * Returns true when a node has a trailing comment, and that comment is on the same line of the
   * beginning line of that node.
   *
   * 

The better detection for nodes that span multiple lines would be to check if a comments * starts on the same line as the node 'ends', but JSC only stores the node's "start" line number, * not the end, thus we cannot check against that at the moment. */ private boolean hasTrailingCommentOnSameLine(Node node) { if (!printNonJSDocComments) { return false; } String nonJSDocCommentString = node.getTrailingNonJSDocCommentString(); if (nonJSDocCommentString.isEmpty()) { return false; } // getLineno() is 1-indexed but getStartPosition().line is 0-indexed. return node.getLineno() == node.getTrailingNonJSDocComment().getStartPosition().line + 1; } protected void printTrailingComment(Node node) { // print any trailing nonJSDoc comment attached to this node if (!printNonJSDocComments) { return; } NonJSDocComment nonJSDocComment = node.getTrailingNonJSDocComment(); if (nonJSDocComment != null) { String nonJSDocCommentString = node.getTrailingNonJSDocCommentString(); if (!nonJSDocCommentString.isEmpty()) { addNonJsDoctrailing(nonJSDocComment, hasTrailingCommentOnSameLine(node)); } } } protected void add(String str) { cc.add(str); } protected void add(Node n) { add(n, Context.OTHER); } private static final QualifiedName JSCOMP_SCOPE = QualifiedName.of("$jscomp.scope"); protected void add(Node node, Context context) { if (!cc.continueProcessing()) { return; } printLeadingCommentsInOrder(node); Token type = node.getToken(); String opstr = NodeUtil.opToStr(type); int childCount = node.getChildCount(); Node first = node.getFirstChild(); Node last = node.getLastChild(); // Handle all binary operators if (opstr != null && first != last) { Preconditions.checkState( childCount == 2, "Bad binary operator \"%s\": expected 2 arguments but got %s", opstr, childCount); int p = precedence(node); // For right-hand-side of operations, only pass context if it's // the IN_FOR_INIT_CLAUSE one. Context rhsContext = getContextForNoInOperator(context); boolean needsParens = (context == Context.START_OF_EXPR || context.atArrowFunctionBody()) && first.isObjectPattern(); if (node.isAssign() && needsParens) { add("("); } if (NodeUtil.isAssignmentOp(node) || type == Token.EXPONENT) { // Assignment operators and '**' are the only right-associative binary operators addExpr(first, p + 1, context); cc.addOp(opstr, true); addExpr(last, p, rhsContext); } else { unrollBinaryOperator(node, type, opstr, context, rhsContext, p, p + 1); } if (node.isAssign() && needsParens) { add(")"); } return; } cc.startSourceMapping(node); switch (type) { case TRY: { checkState(first.getNext().isBlock() && !first.getNext().hasMoreThanOneChild()); checkState(childCount >= 2 && childCount <= 3); add("try"); add(first); // second child contains the catch block, or nothing if there // isn't a catch block Node catchblock = first.getNext().getFirstChild(); if (catchblock != null) { add(catchblock); } if (childCount == 3) { cc.maybeInsertSpace(); add("finally"); add(last); } break; } case CATCH: Preconditions.checkState(childCount == 2, node); cc.maybeInsertSpace(); add("catch"); cc.maybeInsertSpace(); if (!first.isEmpty()) { // optional catch binding add("("); add(first); add(")"); } add(last); break; case THROW: Preconditions.checkState(childCount == 1, node); add("throw"); cc.maybeInsertSpace(); add(first); // Must have a ';' after a throw statement, otherwise safari can't // parse this. cc.endStatement(/*needSemiColon=*/ true, hasTrailingCommentOnSameLine(node)); break; case RETURN: add("return"); if (childCount == 1) { cc.maybeInsertSpace(); if (preserveTypeAnnotations && first.getJSDocInfo() != null) { add("("); add(first); add(")"); } else { add(first); } } else { checkState(childCount == 0, node); } cc.endStatement(hasTrailingCommentOnSameLine(node)); break; case VAR: add("var "); addList(first, false, getContextForNoInOperator(context), ","); if (node.getParent() == null || NodeUtil.isStatement(node)) { cc.endStatement(hasTrailingCommentOnSameLine(node)); } break; case CONST: add("const "); addList(first, false, getContextForNoInOperator(context), ","); if (node.getParent() == null || NodeUtil.isStatement(node)) { cc.endStatement(hasTrailingCommentOnSameLine(node)); } break; case LET: add("let "); addList(first, false, getContextForNoInOperator(context), ","); if (node.getParent() == null || NodeUtil.isStatement(node)) { cc.endStatement(hasTrailingCommentOnSameLine(node)); } break; case LABEL_NAME: Preconditions.checkState(!node.getString().isEmpty(), node); addIdentifier(node.getString()); break; case DESTRUCTURING_LHS: add(first); if (first != last) { checkState(childCount == 2, node); cc.addOp("=", true); addExpr(last, NodeUtil.precedence(Token.ASSIGN), getContextForNoInOperator(context)); } break; case NAME: if (useOriginalName && node.getOriginalName() != null) { addIdentifier(node.getOriginalName()); } else { addIdentifier(node.getString()); } maybeAddOptional(node); maybeAddTypeDecl(node); if (first != null && !first.isEmpty()) { checkState(childCount == 1, node); cc.addOp("=", true); addExpr(first, NodeUtil.precedence(Token.ASSIGN), getContextForNoInOperator(context)); } break; case ARRAYLIT: add("["); addArrayList(first); add("]"); break; case ARRAY_PATTERN: add("["); addArrayList(first); add("]"); maybeAddTypeDecl(node); break; case PARAM_LIST: // If this is the list for a non-TypeScript arrow function with one simple name param. if (node.getParent().isArrowFunction() && node.hasOneChild() && first.isName() && !outputFeatureSet.has(Feature.TYPE_ANNOTATION)) { add(first); } else { add("("); addList(first); add(")"); } break; case DEFAULT_VALUE: add(first); maybeAddTypeDecl(node); cc.addOp("=", true); addExpr(first.getNext(), 1, Context.OTHER); break; case COMMA: Preconditions.checkState(childCount == 2, node); unrollBinaryOperator( node, Token.COMMA, ",", context, getContextForNoInOperator(context), 0, 0); break; case NUMBER: Preconditions.checkState(childCount == 0, node); cc.addNumber(node.getDouble(), node); break; case BIGINT: Preconditions.checkState(childCount == 0, node); cc.add(node.getBigInt() + "n"); break; case TYPEOF: case VOID: case NOT: case BITNOT: case POS: case NEG: { // All of these unary operators are right-associative checkState(childCount == 1, node); cc.addOp(NodeUtil.opToStrNoFail(type), false); addExpr(first, NodeUtil.precedence(type), Context.OTHER); break; } case HOOK: { checkState(childCount == 3, "%s wrong number of children: %s", node, childCount); int p = NodeUtil.precedence(type); Context rhsContext = getContextForNoInOperator(context); addExpr(first, p + 1, context); cc.addOp("?", true); addExpr(first.getNext(), 1, rhsContext); cc.addOp(":", true); addExpr(last, 1, rhsContext); break; } case REGEXP: if (!first.isStringLit() || !last.isStringLit()) { throw new Error("Expected children to be strings"); } String regexp = regexpEscape(first.getString()); // I only use one .add because whitespace matters if (childCount == 2) { add(regexp + last.getString()); } else { checkState(childCount == 1, node); add(regexp); } break; case FUNCTION: { if (node.getClass() != Node.class) { throw new Error("Unexpected Node subclass."); } checkState(childCount == 3, node); if (node.isArrowFunction()) { addArrowFunction(node, first, last, context); } else { addFunction(node, first, last, context); } break; } case ITER_REST: case OBJECT_REST: add("..."); add(first); maybeAddTypeDecl(node); break; case ITER_SPREAD: case OBJECT_SPREAD: add("..."); addExpr(first, NodeUtil.precedence(type), Context.OTHER); break; case EXPORT: add("export"); if (node.getBooleanProp(Node.EXPORT_DEFAULT)) { add("default"); } if (node.getBooleanProp(Node.EXPORT_ALL_FROM)) { add("*"); checkState(first != null && first.isEmpty(), node); } else { add(first); } if (childCount == 2) { add("from"); add(last); } processEnd(first, context); break; case IMPORT: add("import"); Node second = first.getNext(); if (!first.isEmpty()) { add(first); if (!second.isEmpty()) { cc.listSeparator(); } } if (!second.isEmpty()) { add(second); } if (!first.isEmpty() || !second.isEmpty()) { add("from"); } add(last); cc.endStatement(hasTrailingCommentOnSameLine(node)); break; case EXPORT_SPECS: case IMPORT_SPECS: add("{"); for (Node c = first; c != null; c = c.getNext()) { if (c != first) { cc.listSeparator(); } add(c); } add("}"); break; case EXPORT_SPEC: case IMPORT_SPEC: add(first); if (node.isShorthandProperty() && first.getString().equals(last.getString())) { break; } add("as"); add(last); break; case IMPORT_STAR: add("*"); add("as"); add(node.getString()); break; case DYNAMIC_IMPORT: add("import("); addExpr(first, NodeUtil.precedence(type), context); add(")"); break; case IMPORT_META: add("import.meta"); break; // CLASS -> NAME,EXPR|EMPTY,BLOCK case CLASS: { checkState(childCount == 3, node); boolean classNeedsParens = (context == Context.START_OF_EXPR); if (classNeedsParens) { add("("); } Node name = first; Node superClass = first.getNext(); Node members = last; add("class"); if (!name.isEmpty()) { add(name); } maybeAddGenericTypes(first); if (!superClass.isEmpty()) { add("extends"); add(superClass); } Node interfaces = (Node) node.getProp(Node.IMPLEMENTS); if (interfaces != null) { add("implements"); Node child = interfaces.getFirstChild(); add(child); while ((child = child.getNext()) != null) { add(","); cc.maybeInsertSpace(); add(child); } } add(members); cc.endClass(context == Context.STATEMENT); if (classNeedsParens) { add(")"); } } break; case CLASS_MEMBERS: case INTERFACE_MEMBERS: case NAMESPACE_ELEMENTS: cc.beginBlock(); for (Node c = first; c != null; c = c.getNext()) { add(c); processEnd(c, context); cc.endLine(); } cc.endBlock(false); break; case ENUM_MEMBERS: cc.beginBlock(); for (Node c = first; c != null; c = c.getNext()) { add(c); if (c.getNext() != null) { add(","); } cc.endLine(); } cc.endBlock(false); break; case GETTER_DEF: case SETTER_DEF: case MEMBER_FUNCTION_DEF: case MEMBER_VARIABLE_DEF: { checkState( node.getParent().isObjectLit() || node.getParent().isClassMembers() || node.getParent().isInterfaceMembers() || node.getParent().isRecordType() || node.getParent().isIndexSignature()); maybeAddAccessibilityModifier(node); if (node.isStaticMember()) { add("static "); } if (node.isMemberFunctionDef() && node.getFirstChild().isAsyncFunction()) { add("async "); } if (!node.isMemberVariableDef() && node.getFirstChild().isGeneratorFunction()) { checkState(type == Token.MEMBER_FUNCTION_DEF, node); add("*"); } switch (type) { case GETTER_DEF: // Get methods have no parameters. Preconditions.checkState(!first.getSecondChild().hasChildren(), node); add("get "); break; case SETTER_DEF: // Set methods have one parameter. Preconditions.checkState(first.getSecondChild().hasOneChild(), node); add("set "); break; case MEMBER_FUNCTION_DEF: case MEMBER_VARIABLE_DEF: // nothing to do. break; default: break; } // The name is on the GET or SET node. String name = node.getString(); if (node.isMemberVariableDef()) { add(node.getString()); maybeAddOptional(node); maybeAddTypeDecl(node); } else { checkState(childCount == 1, node); checkState(first.isFunction(), first); // The function referenced by the definition should always be unnamed. checkState(first.getFirstChild().getString().isEmpty(), first); Node fn = first; Node parameters = fn.getSecondChild(); Node body = fn.getLastChild(); // Add the property name. if (!node.isQuotedString() && TokenStream.isJSIdentifier(name) && // do not encode literally any non-literal characters that were // Unicode escaped. NodeUtil.isLatin(name)) { add(name); maybeAddGenericTypes(fn.getFirstChild()); } else { // Determine if the string is a simple number. double d = getSimpleNumber(name); if (!Double.isNaN(d)) { cc.addNumber(d, node); } else { addJsString(node); } } maybeAddOptional(fn); add(parameters); maybeAddTypeDecl(fn); add(body); } break; } case MEMBER_FIELD_DEF: case COMPUTED_FIELD_DEF: { checkState(node.getParent().isClassMembers()); if (node.getBooleanProp(Node.STATIC_MEMBER)) { add("static "); } Node init = null; switch (type) { case MEMBER_FIELD_DEF: String propertyName = node.getString(); add(propertyName); init = first; break; case COMPUTED_FIELD_DEF: add("["); // Must use addExpr() with a priority of 1, because comma expressions aren't allowed. // https://www.ecma-international.org/ecma-262/9.0/index.html#prod-ComputedPropertyName addExpr(first, 1, Context.OTHER); add("]"); init = node.getSecondChild(); break; default: break; } if (init != null) { add("="); addExpr(init, 1, Context.OTHER); } add(";"); break; } case SCRIPT: case MODULE_BODY: case BLOCK: case ROOT: { if (node.getClass() != Node.class) { throw new Error("Unexpected Node subclass."); } boolean preserveBlock = node.isBlock() && !node.isSyntheticBlock(); if (preserveBlock) { cc.beginBlock(); } boolean preferLineBreaks = type == Token.SCRIPT || (type == Token.BLOCK && !preserveBlock && node.getParent().isScript()); for (Node c = first; c != null; c = c.getNext()) { add(c, Context.STATEMENT); if (c.isFunction() || c.isClass()) { cc.maybeLineBreak(); } // Prefer to break lines in between top-level statements // because top-level statements are more homogeneous. if (preferLineBreaks) { cc.notePreferredLineBreak(); } } if (preserveBlock) { cc.endBlock(cc.breakAfterBlockFor(node, context == Context.STATEMENT)); } break; } case FOR: Preconditions.checkState(childCount == 4, node); add("for"); cc.maybeInsertSpace(); add("("); if (NodeUtil.isNameDeclaration(first)) { add(first, Context.IN_FOR_INIT_CLAUSE); } else { addExpr(first, 0, Context.IN_FOR_INIT_CLAUSE); } add(";"); if (!first.getNext().isEmpty()) { cc.maybeInsertSpace(); } add(first.getNext()); add(";"); if (!first.getNext().getNext().isEmpty()) { cc.maybeInsertSpace(); } add(first.getNext().getNext()); add(")"); addNonEmptyStatement(last, getContextForNonEmptyExpression(context), false); break; case FOR_IN: Preconditions.checkState(childCount == 3, node); add("for"); cc.maybeInsertSpace(); add("("); add(first); add("in"); add(first.getNext()); add(")"); addNonEmptyStatement(last, getContextForNonEmptyExpression(context), false); break; case FOR_OF: Preconditions.checkState(childCount == 3, node); add("for"); cc.maybeInsertSpace(); add("("); add(first); cc.maybeInsertSpace(); add("of"); cc.maybeInsertSpace(); // the iterable must be an AssignmentExpression addExpr(first.getNext(), NodeUtil.precedence(Token.ASSIGN), Context.OTHER); add(")"); addNonEmptyStatement(last, getContextForNonEmptyExpression(context), false); break; case FOR_AWAIT_OF: Preconditions.checkState(childCount == 3, node); add("for await"); cc.maybeInsertSpace(); add("("); add(first); cc.maybeInsertSpace(); add("of"); cc.maybeInsertSpace(); // the iterable must be an AssignmentExpression addExpr(first.getNext(), NodeUtil.precedence(Token.ASSIGN), Context.OTHER); add(")"); addNonEmptyStatement(last, getContextForNonEmptyExpression(context), false); break; case DO: Preconditions.checkState(childCount == 2, node); add("do"); addNonEmptyStatement(first, Context.OTHER, false); cc.maybeInsertSpace(); add("while"); cc.maybeInsertSpace(); add("("); add(last); add(")"); cc.endStatement(hasTrailingCommentOnSameLine(node)); break; case WHILE: Preconditions.checkState(childCount == 2, node); add("while"); cc.maybeInsertSpace(); add("("); add(first); add(")"); addNonEmptyStatement(last, getContextForNonEmptyExpression(context), false); break; case EMPTY: Preconditions.checkState(childCount == 0, node); break; case OPTCHAIN_GETPROP: { addExpr(first, NodeUtil.precedence(type), context); add(node.isOptionalChainStart() ? "?." : "."); addGetpropIdentifier(node); break; } case GETPROP: { // This attempts to convert rewritten aliased code back to the original code, // such as when using goog.scope(). See ScopedAliases.java for the original code. // NOTE: OPTCHAIN_GETPROP case doesn't need this logic, because it only applies to // qualified names. if (useOriginalName && node.getOriginalName() != null) { // The ScopedAliases pass will convert variable assignments and function declarations // to assignments to GETPROP nodes, like $jscomp.scope.SOME_VAR = 3;. This attempts to // rewrite it back to the original code. if (JSCOMP_SCOPE.matches(node.getFirstChild()) && node.getParent().isAssign()) { add("var "); } addGetpropIdentifier(node); break; } // We need parentheses to distinguish // `a?.b.c` from `(a?.b).c` boolean breakOutOfOptionalChain = NodeUtil.isOptChainNode(first); // `2.toString()` is invalid - it must be `(2).toString()` boolean needsParens = first.isNumber() || breakOutOfOptionalChain; if (needsParens) { add("("); } addExpr(first, NodeUtil.precedence(type), context); if (needsParens) { add(")"); } if (quoteKeywordProperties && TokenStream.isKeyword(node.getString())) { // NOTE: We don't have to worry about quoting keyword properties in the // OPTCHAIN_GETPROP case above, because we only need to quote keywords for // ES3-compatible output. // // Must be a single call to `add` otherwise the generator will add a trailing space. add("[\"" + node.getString() + "\"]"); } else { add("."); addGetpropIdentifier(node); } break; } case OPTCHAIN_GETELEM: { checkState( childCount == 2, "Bad GETELEM node: Expected 2 children but got %s. For node: %s", childCount, node); addExpr(first, NodeUtil.precedence(type), context); if (node.isOptionalChainStart()) { add("?."); } add("["); add(first.getNext()); add("]"); break; } case GETELEM: { checkState( childCount == 2, "Bad GETELEM node: Expected 2 children but got %s. For node: %s", childCount, node); boolean needsParens = NodeUtil.isOptChainNode(first); if (needsParens) { add("("); } addExpr(first, NodeUtil.precedence(type), context); if (needsParens) { add(")"); } add("["); add(first.getNext()); add("]"); break; } case WITH: Preconditions.checkState(childCount == 2, node); add("with("); add(first); add(")"); addNonEmptyStatement(last, getContextForNonEmptyExpression(context), false); break; case INC: case DEC: { checkState(childCount == 1, node); String o = type == Token.INC ? "++" : "--"; boolean postProp = node.getBooleanProp(Node.INCRDECR_PROP); if (postProp) { addExpr(first, NodeUtil.precedence(type), context); cc.addOp(o, false); } else { cc.addOp(o, false); add(first); } break; } case OPTCHAIN_CALL: { // We have two special cases here: // 1) If the left hand side of the call is a direct reference to eval, // then it must have a DIRECT_EVAL annotation. If it does not, then // that means it was originally an indirect call to eval, and that // indirectness must be preserved. // 2) If the left hand side of the call is a property reference, // then the call must not a FREE_CALL annotation. If it does, then // that means it was originally an call without an explicit this and // that must be preserved. if (isIndirectEval(first) || (node.getBooleanProp(Node.FREE_CALL) && NodeUtil.isNormalOrOptChainGet(first))) { add("(0,"); addExpr(first, NodeUtil.precedence(Token.COMMA), Context.OTHER); add(")"); } else { addExpr(first, NodeUtil.precedence(type), context); } Node args = first.getNext(); if (node.isOptionalChainStart()) { add("?."); } add("("); addList(args); add(")"); break; } case CALL: this.addInvocationTarget(node, context); add("("); addList(first.getNext()); add(")"); break; case IF: Preconditions.checkState(childCount == 2 || childCount == 3, node); boolean hasElse = childCount == 3; boolean ambiguousElseClause = context == Context.BEFORE_DANGLING_ELSE && !hasElse; if (ambiguousElseClause) { cc.beginBlock(); } add("if"); cc.maybeInsertSpace(); add("("); add(first); add(")"); if (hasElse) { addNonEmptyStatement(first.getNext(), Context.BEFORE_DANGLING_ELSE, false); cc.maybeInsertSpace(); add("else"); addNonEmptyStatement(last, getContextForNonEmptyExpression(context), false); } else { addNonEmptyStatement(first.getNext(), Context.OTHER, false); } if (ambiguousElseClause) { cc.endBlock(); } break; case NULL: Preconditions.checkState(childCount == 0, node); cc.addConstant("null"); break; case THIS: Preconditions.checkState(childCount == 0, node); add("this"); break; case SUPER: Preconditions.checkState(childCount == 0, node); add("super"); break; case NEW_TARGET: Preconditions.checkState(childCount == 0, node); add("new.target"); break; case YIELD: add("yield"); if (node.isYieldAll()) { checkNotNull(first); add("*"); } if (first != null) { cc.maybeInsertSpace(); addExpr(first, NodeUtil.precedence(type), Context.OTHER); } break; case AWAIT: add("await "); addExpr(first, NodeUtil.precedence(type), Context.OTHER); break; case FALSE: Preconditions.checkState(childCount == 0, node); cc.addConstant("false"); break; case TRUE: Preconditions.checkState(childCount == 0, node); cc.addConstant("true"); break; case CONTINUE: Preconditions.checkState(childCount <= 1, node); add("continue"); if (childCount == 1) { if (!first.isLabelName()) { throw new Error("Unexpected token type. Should be LABEL_NAME."); } add(" "); add(first); } cc.endStatement(hasTrailingCommentOnSameLine(node)); break; case DEBUGGER: Preconditions.checkState(childCount == 0, node); add("debugger"); cc.endStatement(hasTrailingCommentOnSameLine(node)); break; case BREAK: Preconditions.checkState(childCount <= 1, node); add("break"); if (childCount == 1) { if (!first.isLabelName()) { throw new Error("Unexpected token type. Should be LABEL_NAME."); } add(" "); add(first); } cc.endStatement(hasTrailingCommentOnSameLine(node)); break; case EXPR_RESULT: Preconditions.checkState(childCount == 1, node); add(first, Context.START_OF_EXPR); cc.endStatement(hasTrailingCommentOnSameLine(node)); break; case NEW: add("new "); int precedence = NodeUtil.precedence(type); // `new void 0` is a syntax error add parenthese in this case. This is only particularly // interesting for code in dead branches. int precedenceOfFirst = NodeUtil.precedence(first.getToken()); if (precedenceOfFirst == precedence) { precedence = precedence + 1; } // If the first child contains a CALL, then claim higher precedence // to force parentheses. Otherwise, when parsed, NEW will bind to the // first viable parentheses (don't traverse into functions). // Also, NEW requires parentheses around an optional chain callee. // If the first child is an arrow function, then parentheses is needed if (NodeUtil.has(first, Node::isCall, NodeUtil.MATCH_NOT_FUNCTION) || NodeUtil.isOptChainNode(first)) { precedence = NodeUtil.precedence(first.getToken()) + 1; } addExpr(first, precedence, Context.OTHER); // '()' is optional when no arguments are present Node next = first.getNext(); if (next != null) { add("("); addList(next); add(")"); } else { if (cc.shouldPreserveExtras(node)) { add("("); add(")"); } } break; case STRING_KEY: addStringKey(node); break; case STRINGLIT: Preconditions.checkState(childCount == 0, "String node %s may not have children", node); addJsString(node); break; case DELPROP: Preconditions.checkState(childCount == 1, node); add("delete "); add(first); break; case OBJECTLIT: { boolean needsParens = context == Context.START_OF_EXPR || context.atArrowFunctionBody(); if (needsParens) { add("("); } add("{"); for (Node c = first; c != null; c = c.getNext()) { if (c != first) { cc.listSeparator(); } checkState(NodeUtil.isObjLitProperty(c) || c.isSpread(), c); add(c); } if (first != null && node.hasTrailingComma()) { cc.optionalListSeparator(); } add("}"); if (needsParens) { add(")"); } break; } case COMPUTED_PROP: maybeAddAccessibilityModifier(node); if (node.getBooleanProp(Node.STATIC_MEMBER)) { add("static "); } if (node.getBooleanProp(Node.COMPUTED_PROP_GETTER)) { add("get "); } else if (node.getBooleanProp(Node.COMPUTED_PROP_SETTER)) { add("set "); } else if (node.getBooleanProp(Node.COMPUTED_PROP_METHOD)) { if (last.isAsyncFunction()) { add("async"); } if (last.getBooleanProp(Node.GENERATOR_FN)) { add("*"); } } add("["); // Must use addExpr() with a priority of 1, because comma expressions aren't allowed. // https://www.ecma-international.org/ecma-262/9.0/index.html#prod-ComputedPropertyName addExpr(first, 1, Context.OTHER); add("]"); // TODO(martinprobst): There's currently no syntax for properties in object literals that // have type declarations on them (a la `{foo: number: 12}`). This comes up for, e.g., // function parameters with default values. Support when figured out. maybeAddTypeDecl(node); if (node.getBooleanProp(Node.COMPUTED_PROP_METHOD) || node.getBooleanProp(Node.COMPUTED_PROP_GETTER) || node.getBooleanProp(Node.COMPUTED_PROP_SETTER)) { Node function = first.getNext(); Node params = function.getSecondChild(); Node body = function.getLastChild(); add(params); add(body); } else { // This is a field or object literal property. boolean isInClass = node.getParent().isClassMembers(); Node initializer = first.getNext(); if (initializer != null) { // Object literal value. checkState( !isInClass, "initializers should only exist in object literals, not classes"); cc.add(":"); // Must use addExpr() with a priority of 1, because a comma expression here would cause // a syntax error within the object literal. addExpr(initializer, 1, Context.OTHER); } else { // Computed properties must either have an initializer or be computed member-variable // properties that exist for their type declaration. checkState(node.getBooleanProp(Node.COMPUTED_PROP_VARIABLE), node); } } break; case OBJECT_PATTERN: addObjectPattern(node); maybeAddTypeDecl(node); break; case SWITCH: add("switch("); add(first); add(")"); cc.beginBlock(); addAllSiblings(first.getNext()); cc.endBlock(context == Context.STATEMENT); break; case CASE: Preconditions.checkState(childCount == 2, node); add("case "); add(first); addCaseBody(last); break; case DEFAULT_CASE: Preconditions.checkState(childCount == 1, node); add("default"); addCaseBody(first); break; case LABEL: Preconditions.checkState(childCount == 2, node); if (!first.isLabelName()) { throw new Error("Unexpected token type. Should be LABEL_NAME."); } add(first); add(":"); if (!last.isBlock()) { cc.maybeInsertSpace(); } addNonEmptyStatement(last, getContextForNonEmptyExpression(context), true); break; case CAST: if (preserveTypeAnnotations) { add("("); add(first); // drop context because of added parentheses add(")"); } else { add(first, context); // preserve context } break; case TAGGED_TEMPLATELIT: this.addInvocationTarget(node, context); add(first.getNext()); break; case TEMPLATELIT: cc.beginTemplateLit(); for (Node c = first; c != null; c = c.getNext()) { if (c.isTemplateLitString()) { add(escapeUnrecognizedCharacters(c.getRawString())); } else { cc.beginTemplateLitSub(); add(c.getFirstChild(), Context.START_OF_EXPR); cc.endTemplateLitSub(); } } cc.endTemplateLit(); break; // Type Declaration ASTs. case STRING_TYPE: add("string"); break; case BOOLEAN_TYPE: add("boolean"); break; case NUMBER_TYPE: add("number"); break; case ANY_TYPE: add("any"); break; case VOID_TYPE: add("void"); break; case NAMED_TYPE: // Children are a chain of getprop nodes. add(first); break; case ARRAY_TYPE: addExpr(first, NodeUtil.precedence(Token.ARRAY_TYPE), context); add("[]"); break; case FUNCTION_TYPE: Node returnType = first; add("("); addList(first.getNext()); add(")"); cc.addOp("=>", true); add(returnType); break; case UNION_TYPE: addList(first, "|"); break; case RECORD_TYPE: add("{"); addList(first, false, Context.OTHER, ","); add("}"); break; case PARAMETERIZED_TYPE: // First child is the type that's parameterized, later children are the arguments. add(first); add("<"); addList(first.getNext()); add(">"); break; // CLASS -> NAME,EXPR|EMPTY,BLOCK case GENERIC_TYPE_LIST: add("<"); addList(first, false, Context.STATEMENT, ","); add(">"); break; case GENERIC_TYPE: addIdentifier(node.getString()); if (node.hasChildren()) { add("extends"); cc.maybeInsertSpace(); add(node.getFirstChild()); } break; case INTERFACE: { checkState(childCount == 3, node); Node name = first; Node superTypes = first.getNext(); Node members = last; add("interface"); add(name); maybeAddGenericTypes(name); if (!superTypes.isEmpty()) { add("extends"); Node superType = superTypes.getFirstChild(); add(superType); while ((superType = superType.getNext()) != null) { add(","); cc.maybeInsertSpace(); add(superType); } } add(members); } break; case ENUM: { checkState(childCount == 2, node); Node name = first; Node members = last; add("enum"); add(name); add(members); break; } case NAMESPACE: { checkState(childCount == 2, node); Node name = first; Node elements = last; add("namespace"); add(name); add(elements); break; } case TYPE_ALIAS: add("type"); add(node.getString()); cc.addOp("=", true); add(last); cc.endStatement(/* needSemiColon= */ true, hasTrailingCommentOnSameLine(node)); break; case DECLARE: add("declare"); add(first); processEnd(node, context); break; case INDEX_SIGNATURE: add("["); add(first); add("]"); maybeAddTypeDecl(node); cc.endStatement(/* needSemiColon= */ true, hasTrailingCommentOnSameLine(node)); break; case CALL_SIGNATURE: if (node.getBooleanProp(Node.CONSTRUCT_SIGNATURE)) { add("new "); } maybeAddGenericTypes(node); add(first); maybeAddTypeDecl(node); cc.endStatement(/* needSemiColon= */ true, hasTrailingCommentOnSameLine(node)); break; default: throw new IllegalStateException("Unknown token " + type + "\n" + node.toStringTree()); } printTrailingComment(node); cc.endSourceMapping(node); } private void addIdentifier(String identifier) { cc.addIdentifier(identifierEscape(identifier)); } private void addGetpropIdentifier(Node getprop) { cc.startSourceMapping(getprop); addIdentifier(getprop.getString()); cc.endSourceMapping(getprop); } private int precedence(Node n) { if (n.isCast()) { return precedence(n.getFirstChild()); } return NodeUtil.precedence(n.getToken()); } /** * We have two special cases here: * *

    *
  • If the left hand side of the call is a direct reference to eval, then it must have a * DIRECT_EVAL annotation. If it does not, then that means it was originally an indirect * call to eval, and that indirectness must be preserved. *
  • If the left hand side of the call is a property reference, then the call must not a * FREE_CALL annotation. If it does, then that means it was originally an call without an * explicit this and that must be preserved. */ private void addInvocationTarget(Node node, Context context) { Node first = node.getFirstChild(); boolean needsParens = NodeUtil.isOptChainNode(first); if (isIndirectEval(first) || (node.getBooleanProp(Node.FREE_CALL) && NodeUtil.isNormalOrOptChainGet(first))) { add("(0,"); addExpr(first, NodeUtil.precedence(Token.COMMA), Context.OTHER); add(")"); } else { if (needsParens) { add("("); } addExpr(first, NodeUtil.precedence(node.getToken()), context); if (needsParens) { add(")"); } } } private static boolean arrowFunctionNeedsParens(Node n) { Node parent = n.getParent(); // Once you cut through the layers of non-terminals used to define operator precedence, // you can see the following are true. // (Read => as "may expand to" and "!=>" as "may not expand to") // // 1. You can substitute an ArrowFunction into rules where an Expression or // AssignmentExpression is required, because // Expression => AssignmentExpression => ArrowFunction // // 2. However, most operators act on LeftHandSideExpression, CallExpression, or // MemberExpression. None of these expand to ArrowFunction. // // 3. CallExpression cannot expand to an ArrowFunction at all, because all of its expansions // produce multiple symbols and none can be logically equivalent to ArrowFunction. // // 4. LeftHandSideExpression and MemberExpression may be replaced with an ArrowFunction in // parentheses, because: // LeftHandSideExpression => MemberExpression => PrimaryExpression // PrimaryExpression => '(' Expression ')' => '(' ArrowFunction ')' if (parent == null) { return false; } else if (NodeUtil.isBinaryOperator(parent) || NodeUtil.isUnaryOperator(parent) || NodeUtil.isUpdateOperator(parent) || parent.isTaggedTemplateLit() || parent.isGetProp() || parent.isOptChainGetProp()) { // LeftHandSideExpression OP LeftHandSideExpression // OP LeftHandSideExpression | LeftHandSideExpression OP // MemberExpression TemplateLiteral // MemberExpression '.' IdentifierName return true; } else if (parent.isGetElem() || parent.isCall() || parent.isHook() || parent.isOptChainGetElem() || parent.isOptChainCall() || parent.isNew()) { // MemberExpression '[' Expression ']' // MemberFunction '(' AssignmentExpressionList ')' // LeftHandSideExpression ? AssignmentExpression : AssignmentExpression return isFirstChild(n); } else { // All other cases are either illegal (e.g. because you cannot assign a value to an // ArrowFunction) or do not require parens. return false; } } private static boolean isFirstChild(Node n) { Node parent = n.getParent(); return parent != null && n == parent.getFirstChild(); } private void addArrowFunction(Node n, Node first, Node last, Context context) { checkState(first.getString().isEmpty(), first); boolean funcNeedsParens = arrowFunctionNeedsParens(n); if (funcNeedsParens) { add("("); } maybeAddGenericTypes(first); if (n.isAsyncFunction()) { add("async"); } add(first.getNext()); // param list maybeAddTypeDecl(n); cc.addOp("=>", true); if (last.isBlock()) { add(last); } else { // This is a hack. Arrow functions have no token type, but // blockless arrow function bodies have lower precedence than anything other than commas. addExpr(last, NodeUtil.precedence(Token.COMMA) + 1, getContextForArrowFunctionBody(context)); } cc.endFunction(context == Context.STATEMENT); if (funcNeedsParens) { add(")"); } } private void addFunction(Node n, Node first, Node last, Context context) { boolean funcNeedsParens = (context == Context.START_OF_EXPR); if (funcNeedsParens) { add("("); } add(n.isAsyncFunction() ? "async function" : "function"); if (n.isGeneratorFunction()) { add("*"); if (!first.getString().isEmpty()) { cc.maybeInsertSpace(); } } add(first); maybeAddGenericTypes(first); add(first.getNext()); // param list maybeAddTypeDecl(n); add(last); cc.endFunction(context == Context.STATEMENT); if (funcNeedsParens) { add(")"); } } private void maybeAddAccessibilityModifier(Node n) { Visibility access = (Visibility) n.getProp(Node.ACCESS_MODIFIER); if (access != null) { add(Ascii.toLowerCase(access.toString()) + " "); } } private void maybeAddTypeDecl(Node n) { if (n.getDeclaredTypeExpression() != null) { add(":"); cc.maybeInsertSpace(); add(n.getDeclaredTypeExpression()); } } private void maybeAddGenericTypes(Node n) { Node generics = (Node) n.getProp(Node.GENERIC_TYPE_LIST); if (generics != null) { add(generics); } } private void maybeAddOptional(Node n) { if (n.getBooleanProp(Node.OPT_ES6_TYPED)) { add("?"); } } /** * We could use addList recursively here, but sometimes we produce very deeply nested operators * and run out of stack space, so we just unroll the recursion when possible. * *

    We assume nodes are left-recursive. */ private void unrollBinaryOperator( Node n, Token op, String opStr, Context context, Context rhsContext, int leftPrecedence, int rightPrecedence) { Node firstNonOperator = n.getFirstChild(); while (firstNonOperator.getToken() == op) { firstNonOperator = firstNonOperator.getFirstChild(); } addExpr(firstNonOperator, leftPrecedence, context); Node current = firstNonOperator; do { current = current.getParent(); cc.addOp(opStr, true); addExpr(current.getSecondChild(), rightPrecedence, rhsContext); } while (current != n); } static boolean isSimpleNumber(String s) { int len = s.length(); if (len == 0) { return false; } for (int index = 0; index < len; index++) { char c = s.charAt(index); if (c < '0' || c > '9') { return false; } } return len == 1 || s.charAt(0) != '0'; } static double getSimpleNumber(String s) { if (isSimpleNumber(s)) { try { long l = Long.parseLong(s); if (l <= NodeUtil.MAX_POSITIVE_INTEGER_NUMBER) { return l; } } catch (NumberFormatException e) { // The number was too long to parse. Fall through to NaN. } } return Double.NaN; } /** * @return Whether the name is an indirect eval. */ private static boolean isIndirectEval(Node n) { return n.isName() && "eval".equals(n.getString()) && !n.getBooleanProp(Node.DIRECT_EVAL); } /** * Adds a block or expression, substituting a VOID with an empty statement. This is used for "for * (...);" and "if (...);" type statements. * * @param n The node to print. * @param context The context to determine how the node should be printed. */ private void addNonEmptyStatement(Node n, Context context, boolean allowNonBlockChild) { Node nodeToProcess = n; if (!allowNonBlockChild && !n.isBlock()) { throw new Error("Missing BLOCK child."); } // Strip unneeded blocks, that is blocks with <2 children unless // the CodePrinter specifically wants to keep them. if (n.isBlock()) { int count = getNonEmptyChildCount(n, 2); if (count == 0) { if (cc.shouldPreserveExtras(n)) { cc.beginBlock(); printTrailingComment(n); cc.endBlock(cc.breakAfterBlockFor(n, context == Context.STATEMENT)); } else { printTrailingComment(n); cc.endStatement(/* needSemiColon= */ true, /*hasTrailingCommentOnSameLine=*/ false); } return; } if (count == 1) { // Preserve the block only if needed or requested. // 'let', 'const', etc are not allowed by themselves in "if" and other // structures. Also, hack around a IE6/7 browser bug that needs a block around DOs. Node firstAndOnlyChild = getFirstNonEmptyChild(n); boolean alwaysWrapInBlock = cc.shouldPreserveExtras(n); if (alwaysWrapInBlock || isBlockDeclOrDo(firstAndOnlyChild)) { cc.beginBlock(); add(firstAndOnlyChild, Context.STATEMENT); printTrailingComment(n); cc.maybeLineBreak(); cc.endBlock(cc.breakAfterBlockFor(n, context == Context.STATEMENT)); return; } else { // Continue with the only child. nodeToProcess = firstAndOnlyChild; } } } if (nodeToProcess.isEmpty()) { printTrailingComment(n); cc.endStatement(/*needSemicolon=*/ true, /*hasTrailingCommentOnSameLine=*/ false); } else { add(nodeToProcess, context); printTrailingComment(n); } } /** * @return Whether the Node is a DO or a declaration that is only allowed in restricted contexts. */ private static boolean isBlockDeclOrDo(Node n) { if (n.isLabel()) { Node labeledStatement = n.getLastChild(); if (!labeledStatement.isBlock()) { return isBlockDeclOrDo(labeledStatement); } else { // For labels with block children, we need to ensure that a // labeled FUNCTION or DO isn't generated when extraneous BLOCKs // are skipped. if (getNonEmptyChildCount(n, 2) == 1) { return isBlockDeclOrDo(getFirstNonEmptyChild(n)); } else { // Either a empty statement or an block with more than one child, // way it isn't a FUNCTION or DO. return false; } } } else { switch (n.getToken()) { case LET: case CONST: case FUNCTION: case CLASS: case DO: return true; default: return false; } } } private void addExpr(Node n, int minPrecedence, Context context) { if (opRequiresParentheses(n, minPrecedence, context)) { add("("); add(n, Context.OTHER); add(")"); } else { add(n, context); } } private boolean opRequiresParentheses(Node n, int minPrecedence, Context context) { if (context.inForInInitClause() && n.isIn()) { // make sure this operator 'in' isn't confused with the for-loop 'in' return true; } else if (NodeUtil.isUnaryOperator(n) && isFirstOperandOfExponentiationExpression(n)) { // Unary operators are higher precedence than '**', but // ExponentiationExpression cannot expand to // UnaryExpression ** ExponentiationExpression return true; } else if (isLogicalANDorLogicalORChildOfNullishCoalesce(n) || isNullishCoalesceChildOfLogicalANDorLogicalOR(n)) { // precedence is not enough here since using && or || with ?? without parentheses // is a syntax error as ?? expands directly to | return true; } else { return precedence(n) < minPrecedence; } } private static boolean isLogicalANDorLogicalORChildOfNullishCoalesce(Node n) { Node parent = n.getParent(); boolean logicalANDorLogicalOR = n.isAnd() || n.isOr(); boolean childOfNullishCoalesce = parent != null && parent.isNullishCoalesce(); return logicalANDorLogicalOR && childOfNullishCoalesce; } private static boolean isNullishCoalesceChildOfLogicalANDorLogicalOR(Node n) { Node parent = n.getParent(); boolean childOfLogicalANDorLogicalOR = parent != null && (parent.isAnd() || parent.isOr()); return n.isNullishCoalesce() && childOfLogicalANDorLogicalOR; } private boolean isFirstOperandOfExponentiationExpression(Node n) { Node parent = n.getParent(); return parent != null && parent.getToken() == Token.EXPONENT && parent.getFirstChild() == n; } void addList(Node firstInList) { addList(firstInList, true, Context.OTHER, ","); } void addList(Node firstInList, String separator) { addList(firstInList, true, Context.OTHER, separator); } void addList( Node firstInList, boolean isArrayOrFunctionArgument, Context lhsContext, String separator) { if (firstInList == null) { return; } for (Node n = firstInList; n != null; n = n.getNext()) { boolean isFirst = n == firstInList; int minPrecedence = isArrayOrFunctionArgument ? 1 : 0; if (isFirst) { addExpr(n, minPrecedence, lhsContext); } else { cc.addOp(separator, true); addExpr(n, minPrecedence, getContextForNoInOperator(lhsContext)); } } if (isArrayOrFunctionArgument && checkNotNull(firstInList.getParent()).hasTrailingComma()) { cc.optionalListSeparator(); } } void addStringKey(Node n) { String key = n.getString(); // Object literal property names don't have to be quoted if they are not JavaScript keywords. boolean mustBeQuoted = n.isQuotedString() || (quoteKeywordProperties && TokenStream.isKeyword(key)) || !TokenStream.isJSIdentifier(key) // do not encode literally any non-literal characters that were Unicode escaped. || !NodeUtil.isLatin(key); if (!mustBeQuoted) { // Check if the property is eligible to be printed as shorthand. if (n.isShorthandProperty()) { Node child = n.getFirstChild(); if (child.matchesQualifiedName(key) || (child.isDefaultValue() && child.getFirstChild().matchesQualifiedName(key))) { add(child); return; } } add(key); } else { // Determine if the string is a simple number. double d = getSimpleNumber(key); if (!Double.isNaN(d)) { cc.addNumber(d, n); } else { addJsString(n); } } if (n.hasChildren()) { // NOTE: the only time a STRING_KEY node does *not* have children is when it's // inside a TypeScript enum. We should change these to their own ENUM_KEY token // so that the bifurcating logic can be removed from STRING_KEY. add(":"); addExpr(n.getFirstChild(), 1, Context.OTHER); } } void addObjectPattern(Node n) { add("{"); for (Node child = n.getFirstChild(); child != null; child = child.getNext()) { if (child != n.getFirstChild()) { cc.listSeparator(); } add(child); } add("}"); } /** * Adds a comma-separated list as is specified by an ARRAYLIT node. * * @param firstInList The first in the node list (chained through the next property). */ void addArrayList(@Nullable Node firstInList) { if (firstInList == null) { return; } boolean lastWasEmpty = false; for (Node n = firstInList; n != null; n = n.getNext()) { if (n != firstInList) { cc.listSeparator(); } addExpr(n, 1, Context.OTHER); lastWasEmpty = n.isEmpty(); } if (lastWasEmpty) { cc.listSeparator(); } else if (firstInList.getParent().hasTrailingComma()) { cc.optionalListSeparator(); } } void addCaseBody(Node caseBody) { checkState(caseBody.isBlock(), caseBody); cc.beginCaseBody(); addAllSiblings(caseBody.getFirstChild()); cc.endCaseBody(); } void addAllSiblings(Node n) { for (Node c = n; c != null; c = c.getNext()) { add(c); } } private void addNonJsDoc_nonTrailing(Node node, NonJSDocComment nonJSDocComment) { String content = nonJSDocComment.getCommentString(); SourcePosition commentEndPosition = nonJSDocComment.getEndPosition(); int nodeLineNumber = node.getLineno() - 1; // source lines are 1-indexed if (nonJSDocComment.isEndingAsLineComment()) { // Non trailing line comments can not be on the same line as the node. checkState( commentEndPosition.line < nodeLineNumber, "Non trailing line comments can not be on the same line as the node."); add(content + "\n"); } else { if (nodeLineNumber == commentEndPosition.line) { // e.g. ``` /* comment */ let x; ``` add(content + " "); } else { // e.g. // ``` // /* comment */ // let x; // ``` add(content + "\n"); } } } private void addNonJsDoctrailing(NonJSDocComment nonJSDocComment, boolean sameLine) { String content = nonJSDocComment.getCommentString(); if (nonJSDocComment.isEndingAsLineComment()) { // Trailing line comments *must* end with a `\n`. E.g.. `let x; //comment\n` add(" " + content); if (sameLine) { cc.startNewLine(); } } else { if (nonJSDocComment.isInline()) { // e.g. `foo(x /*comment*/);` is inline add(" " + content); } else { // e.g. `let x; /*comment*/` is non-inline add(" " + content); if (sameLine) { cc.startNewLine(); } } } } /** Outputs a JS string, using the optimal (single/double) quote character */ private void addJsString(Node n) { add(jsString(n.getString())); } private String jsString(String s) { int singleq = 0; int doubleq = 0; // could count the quotes and pick the optimal quote character for (int i = 0; i < s.length(); i++) { switch (s.charAt(i)) { case '"': doubleq++; break; case '\'': singleq++; break; default: // skip non-quote characters } } String doublequote; String singlequote; char quote; if (preferSingleQuotes ? (singleq <= doubleq) : (singleq < doubleq)) { // more double quotes so enclose in single quotes. quote = '\''; doublequote = "\""; singlequote = "\\\'"; } else { // more single quotes so escape the doubles quote = '\"'; doublequote = "\\\""; singlequote = "\'"; } return quote + strEscape(s, doublequote, singlequote, "`", "\\\\", "$", false) + quote; } /** Escapes regular expression */ String regexpEscape(String s) { return '/' + strEscape(s, "\"", "'", "`", "\\", "$", true) + '/'; } /** Helper to escape JavaScript string as well as regular expression */ private String strEscape( String s, String doublequoteEscape, String singlequoteEscape, String backtickEscape, String backslashEscape, String dollarEscape, boolean isRegexp) { StringBuilder sb = new StringBuilder(s.length() + 2); for (int i = 0; i < s.length(); i++) { char c = s.charAt(i); switch (c) { case '\0': sb.append("\\x00"); break; case '\u000B': if (!isRegexp) { sb.append("\\v"); } else { sb.append("\\x0B"); } break; // From the SingleEscapeCharacter grammar production. case '\b': sb.append("\\b"); break; case '\f': sb.append("\\f"); break; case '\n': sb.append("\\n"); break; case '\r': sb.append("\\r"); break; case '\t': sb.append("\\t"); break; case '\\': sb.append(backslashEscape); break; case '\"': sb.append(doublequoteEscape); break; case '\'': sb.append(singlequoteEscape); break; case '$': sb.append(dollarEscape); break; case '`': sb.append(backtickEscape); break; case '=': // '=' is a syntactically significant regexp character. if (trustedStrings || isRegexp) { sb.append(c); } else { sb.append("\\x3d"); } break; case '&': if (trustedStrings || isRegexp) { sb.append(c); } else { sb.append("\\x26"); } break; case '>': if (!trustedStrings && !isRegexp) { sb.append(GT_ESCAPED); break; } // Break --> into --\> or ]]> into ]]\> // // This is just to prevent developers from shooting themselves in the // foot, and does not provide the level of security that you get // with trustedString == false. if (i >= 2 && ((s.charAt(i - 1) == '-' && s.charAt(i - 2) == '-') || (s.charAt(i - 1) == ']' && s.charAt(i - 2) == ']'))) { sb.append(GT_ESCAPED); } else { sb.append(c); } break; case '<': if (!trustedStrings && !isRegexp) { sb.append(LT_ESCAPED); break; } // Break 0x1f && c < 0x7f)) { // If we're given an outputCharsetEncoder, then check if the character can be // represented in this character set. If no charsetEncoder provided - pass straight // Latin characters through, and escape the rest. Doing the explicit character check is // measurably faster than using the CharsetEncoder. sb.append(c); } else { // Other characters can be misinterpreted by some JS parsers, // or perhaps mangled by proxies along the way, // so we play it safe and Unicode escape them. Util.appendHexJavaScriptRepresentation(sb, c); } } } return sb.toString(); } /** * Helper to escape the characters that might be misinterpreted * * @param s the string to modify * @return the string with unrecognizable characters escaped. */ private String escapeUnrecognizedCharacters(String s) { // TODO(yitingwang) Move this method to a suitable place StringBuilder sb = new StringBuilder(); for (int i = 0; i < s.length(); i++) { char c = s.charAt(i); switch (c) { // From the SingleEscapeCharacter grammar production. case '\b': case '\f': case '\n': case '\r': case '\t': case '\\': case '\"': case '\'': case '$': case '`': case '\u2028': case '\u2029': sb.append(c); break; default: if ((outputCharsetEncoder != null && outputCharsetEncoder.canEncode(c)) || (c > 0x1f && c < 0x7f)) { // If we're given an outputCharsetEncoder, then check if the character can be // represented in this character set. If no charsetEncoder provided - pass straight // Latin characters through, and escape the rest. Doing the explicit character check is // measurably faster than using the CharsetEncoder. sb.append(c); } else { // Other characters can be misinterpreted by some JS parsers, // or perhaps mangled by proxies along the way, // so we play it safe and Unicode escape them. Util.appendHexJavaScriptRepresentation(sb, c); } } } return sb.toString(); } static String identifierEscape(String s) { // First check if escaping is needed at all -- in most cases it isn't. if (NodeUtil.isLatin(s)) { return s; } // Now going through the string to escape non-Latin characters if needed. StringBuilder sb = new StringBuilder(); for (int i = 0; i < s.length(); i++) { char c = s.charAt(i); // Identifiers should always go to Latin1/ ASCII characters because // different browser's rules for valid identifier characters are // crazy. if (c > 0x1F && c < 0x7F) { sb.append(c); } else { Util.appendHexJavaScriptRepresentation(sb, c); } } return sb.toString(); } /** * @param maxCount The maximum number of children to look for. * @return The number of children of this node that are non empty up to maxCount. */ private static int getNonEmptyChildCount(Node n, int maxCount) { int i = 0; Node c = n.getFirstChild(); for (; c != null && i < maxCount; c = c.getNext()) { if (c.isBlock()) { i += getNonEmptyChildCount(c, maxCount - i); } else if (!c.isEmpty()) { i++; } } return i; } /** Gets the first non-empty child of the given node. */ private static Node getFirstNonEmptyChild(Node n) { for (Node c = n.getFirstChild(); c != null; c = c.getNext()) { if (c.isBlock()) { Node result = getFirstNonEmptyChild(c); if (result != null) { return result; } } else if (!c.isEmpty()) { return c; } } return null; } /** * Information on the current context. Used for disambiguating special cases. For example, a "{" * could indicate the start of an object literal or a block, depending on the current context. */ public enum Context { STATEMENT, BEFORE_DANGLING_ELSE, // a hack to resolve the else-clause ambiguity START_OF_EXPR, // Are we inside the init clause of a for loop? If so, the containing // expression can't contain an in operator. Pass this context flag down // until we reach expressions which no longer have the limitation. IN_FOR_INIT_CLAUSE( /** inForInitClause */ true, /** at start of arrow fn */ false), // Handle object literals at the start of a non-block arrow function body. // This is only important when the first token after the "=>" is "{". START_OF_ARROW_FN_BODY( /** inForInitClause */ false, /** at start of arrow fn */ true), START_OF_ARROW_FN_IN_FOR_INIT( /** inForInitClause */ true, /** atArrowFunctionBody */ true), OTHER; // nothing special to watch out for. // The following two cases are independent, unlike the other enum states, so we have separate // booleans for them. private final boolean inForInitClause; private final boolean atArrowFnBody; Context() { this(false, false); } Context(boolean inForInitClause, boolean atStartOfArrowFnBody) { this.inForInitClause = inForInitClause; this.atArrowFnBody = atStartOfArrowFnBody; } public boolean inForInInitClause() { return inForInitClause; } public boolean atArrowFunctionBody() { return atArrowFnBody; } } private static Context getContextForNonEmptyExpression(Context currentContext) { return currentContext == Context.BEFORE_DANGLING_ELSE ? Context.BEFORE_DANGLING_ELSE : Context.OTHER; } /** * If we're in a IN_FOR_INIT_CLAUSE, we can't permit in operators in the expression. Pass on the * IN_FOR_INIT_CLAUSE flag through subexpressions. */ private static Context getContextForNoInOperator(Context context) { return (context.inForInInitClause() ? context : Context.OTHER); } /** * If we're at the start of an arrow function body, we need parentheses around object literals and * object patterns. We also must also pass the IN_FOR_INIT_CLAUSE flag into subexpressions. */ private static Context getContextForArrowFunctionBody(Context context) { return context.inForInInitClause() ? Context.START_OF_ARROW_FN_IN_FOR_INIT : Context.START_OF_ARROW_FN_BODY; } private void processEnd(Node n, Context context) { switch (n.getToken()) { case CLASS: case INTERFACE: case ENUM: case NAMESPACE: cc.endClass(context == Context.STATEMENT); break; case FUNCTION: if (n.getLastChild().isEmpty()) { cc.endStatement(/* needSemiColon= */ true, /*hasTrailingCommentOnSameLine=*/ false); } else { cc.endFunction(context == Context.STATEMENT); } break; case DECLARE: if (n.getParent().getToken() != Token.NAMESPACE_ELEMENTS) { processEnd(n.getFirstChild(), context); } break; case EXPORT: if (n.getParent().getToken() != Token.NAMESPACE_ELEMENTS && n.getFirstChild().getToken() != Token.DECLARE) { processEnd(n.getFirstChild(), context); } break; case COMPUTED_PROP: if (n.hasOneChild()) { cc.endStatement(/* needSemiColon= */ true, /*hasTrailingCommentOnSameLine=*/ false); } break; case MEMBER_FUNCTION_DEF: case GETTER_DEF: case SETTER_DEF: if (n.getFirstChild().getLastChild().isEmpty()) { cc.endStatement(/* needSemiColon= */ true, /*hasTrailingCommentOnSameLine=*/ false); } break; case MEMBER_VARIABLE_DEF: cc.endStatement(/* needSemiColon= */ true, /*hasTrailingCommentOnSameLine=*/ false); break; default: if (context == Context.STATEMENT) { cc.endStatement(/*hasTrailingCommentOnSameLine=*/ false); } } } }





© 2015 - 2024 Weber Informatics LLC | Privacy Policy