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

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

/*
 * 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.checkState;

import com.google.errorprone.annotations.ForOverride;
import com.google.javascript.rhino.Node;

/**
 * Abstracted consumer of the CodeGenerator output.
 *
 * @see CodeGenerator
 * @see CodePrinter
 * @see InlineCostEstimator
 */
public abstract class CodeConsumer {

  boolean statementNeedsEnded = false;
  boolean statementStarted = false;
  boolean sawFunction = false;

  // State tracking for template literals. Remember that template literal substitutions can contain
  // additional template literals.
  private int templateLitDepth = 0;
  private int templateLitSubDepth = 0;

  /**
   * Starts the source mapping for the given
   * node at the current position.
   */
  void startSourceMapping(Node node) {
  }

  /**
   * Finishes the source mapping for the given
   * node at the current position.
   */
  void endSourceMapping(Node node) {
  }

  /**
   * Provides a means of interrupting the CodeGenerator. Derived classes should
   * return false to stop further processing.
   */
  boolean continueProcessing() {
    return true;
  }

  /**
   * Retrieve the last character of the last string sent to append.
   */
  abstract char getLastChar();

  /**
   * Appends a (possibly multiline) string to the code, keeping track of the current line length and
   * count.
   *
   * 

Clients should use this method rather than {@link #append()}. It sanitizes the input to * {@link #append()}. */ final void add(String newcode) { maybeEndStatement(); if (newcode.isEmpty()) { return; } char c = newcode.charAt(0); if ((isWordChar(c) || c == '\\') && isWordChar(getLastChar())) { // need space to separate. This is not pretty printing. // For example: "return foo;" append(" "); } else if (c == '/' && getLastChar() == '/') { // Do not allow a forward slash to appear after a DIV. // For example, // REGEXP DIV REGEXP // is valid and should print like // / // / / append(" "); } else if ((c == '"' || c == '\'') && isWordChar(getLastChar())) { maybeInsertSpace(); } // Iterate through the new code and add each contained line, followed by a break. Remember that // the string may start and end in the middle of a line. We do this rather primitively because // this method is called frequently and most invocations have only one line. // TODO(nickreid): There are other possible newline characters recognized by the JS spec. We // should also be considering them. int startOfLine = 0; int endOfLine = newcode.indexOf('\n'); while (endOfLine >= 0) { if (endOfLine > startOfLine) { // Append line only if it is non-empty. append(newcode.substring(startOfLine, endOfLine)); } // Breaking is non-optional. Newlines added this way must be preserved (e.g. newlines in // template literals). startNewLine(); startOfLine = endOfLine + 1; // Jump over the newline char. endOfLine = newcode.indexOf('\n', startOfLine); } if (newcode.length() > startOfLine) { // Append line only if it is non-empty. append(newcode.substring(startOfLine)); // Append the last or only line without breaking. } } /** * Appends a string to the code, keeping track of the current line length. * *

Clients should not call this method directly, but instead call add {@link #add()}. * *

The string must be a complete token; partial strings or partial regexes will run the risk of * being split across lines. * *

Implementations of this method need not consider newline characters in {@code str}. Such * characters should either be ignored or cause an {@link Exception}. */ @ForOverride abstract void append(String str); void addIdentifier(String identifier) { add(identifier); } void appendBlockStart() { append("{"); } void appendBlockEnd() { append("}"); } void startNewLine() { } void maybeLineBreak() { maybeCutLine(); } void maybeCutLine() { } void endLine() { } void notePreferredLineBreak() { } void beginBlock() { if (statementNeedsEnded) { append(";"); maybeLineBreak(); } appendBlockStart(); endLine(); statementNeedsEnded = false; } void endBlock() { endBlock(false); } void endBlock(boolean shouldEndLine) { appendBlockEnd(); if (shouldEndLine) { endLine(); } statementNeedsEnded = false; } void listSeparator() { add(","); maybeLineBreak(); } /** * Indicates the end of a statement and a ';' may need to be added. * But we don't add it now, in case we're at the end of a block (in which * case we don't have to add the ';'). * See maybeEndStatement() */ void endStatement() { endStatement(false); } void endStatement(boolean needSemiColon) { if (needSemiColon) { append(";"); maybeLineBreak(); statementNeedsEnded = false; } else if (statementStarted) { statementNeedsEnded = true; } } /** * This is to be called when we're in a statement. If the prev statement * needs to be ended, add a ';'. */ void maybeEndStatement() { // Add a ';' if we need to. if (statementNeedsEnded) { append(";"); maybeLineBreak(); endLine(); statementNeedsEnded = false; } statementStarted = true; } void endFunction(boolean statementContext) { sawFunction = true; if (statementContext) { endLine(); } } void endClass(boolean statementContext) { if (statementContext) { endLine(); } } void beginCaseBody() { append(":"); } void endCaseBody() { } final void beginTemplateLit() { checkState(templateLitDepth == templateLitSubDepth); maybeEndStatement(); append("`"); templateLitDepth++; } final void beginTemplateLitSub() { // This method pair exists because '$' behaves differently inside template literals. We want to // append it without sanitizing in the generic way {@link #add()} would. checkState(isInTemplateLiteral()); append("${"); templateLitSubDepth++; } final void endTemplateLitSub() { // This method pair exists because '$' behaves differently inside template literals. We want to // append it without sanitizing in the generic way {@link #add()} would. checkState(templateLitSubDepth > 0); checkState(templateLitDepth == templateLitSubDepth); append("}"); templateLitSubDepth--; } final void endTemplateLit() { checkState(templateLitDepth > 0); checkState(isInTemplateLiteral()); append("`"); templateLitDepth--; } final boolean isInTemplateLiteral() { // We're inside a template literal but not a substitution within that literal. return templateLitDepth == templateLitSubDepth + 1; } void appendOp(String op, boolean binOp) { append(op); } void addOp(String op, boolean binOp) { maybeEndStatement(); char first = op.charAt(0); char prev = getLastChar(); if ((first == '+' || first == '-') && prev == first) { // This is not pretty printing. This is to prevent misparsing of // things like "x + ++y" or "x++ + ++y" append(" "); } else if (Character.isLetter(first) && isWordChar(prev)) { // Make sure there is a space after e.g. instanceof , typeof append(" "); } else if (prev == '-' && first == '>' || prev == '<' && first == '!') { // Make sure that we don't emit "" append(" "); } // Allow formatting around the operator. appendOp(op, binOp); // Line breaking after an operator is always safe. Line breaking before an // operator on the other hand is not. We only line break after a bin op // because it looks strange. if (binOp) { maybeCutLine(); } } void addNumber(double x, Node n) { // This is not pretty printing. This is to prevent misparsing of x- -4 as // x--4 (which is a syntax error). char prev = getLastChar(); boolean negativeZero = isNegativeZero(x); if ((x < 0 || negativeZero) && prev == '-') { add(" "); } if (negativeZero) { addConstant("-0"); } else if ((long) x == x) { long value = (long) x; long mantissa = value; int exp = 0; if (Math.abs(x) >= 100) { while (mantissa / 10 * ((long) Math.pow(10, exp + 1)) == value) { mantissa /= 10; exp++; } } if (exp > 2) { addConstant(mantissa + "E" + exp); } else { long valueAbs = Math.abs(value); if (valueAbs > 1000000000000L && // Values <1E12 are shorter in decimal Long.toHexString(valueAbs).length() + 2 < Long.toString(valueAbs).length()) { addConstant((value < 0 ? "-" : "") + "0x" + Long.toHexString(valueAbs)); } else { addConstant(Long.toString(value)); } } } else { addConstant(String.valueOf(x).replace(".0E", "E").replaceFirst( "^(-?)0\\.", "$1.")); } } void addConstant(String newcode) { add(newcode); } static boolean isNegativeZero(double x) { return x == 0.0 && 1 / x < 0; } static boolean isWordChar(char ch) { return (ch == '_' || ch == '$' || Character.isLetterOrDigit(ch)); } /** * If the body of a for loop or the then clause of an if statement has * a single statement, should it be wrapped in a block? Doing so can * help when pretty-printing the code, and permits putting a debugging * breakpoint on the statement inside the condition. * * @return {@boolean true} if such expressions should be wrapped */ boolean shouldPreserveExtraBlocks() { return false; } /** * Allows a consumer to insert spaces in locations where it is unnecessary * but may improve the readability of the code. This will be called in such * places as after a statement and before opening parentheses, or after the * end of a if block before the start of an else block. */ void maybeInsertSpace() {} /** * @return Whether the a line break can be added after the specified BLOCK. */ boolean breakAfterBlockFor(Node n, boolean statementContext) { return statementContext; } /** Called when we're at the end of a file. */ void endFile() {} }





© 2015 - 2024 Weber Informatics LLC | Privacy Policy