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

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

Go to download

Closure Compiler is a JavaScript optimizing compiler. It parses your JavaScript, analyzes it, removes dead code and rewrites and minimizes what's left. It also checks syntax, variable references, and types, and warns about common JavaScript pitfalls. It is used in many of Google's JavaScript apps, including Gmail, Google Web Search, Google Maps, and Google Docs.

There is a newer version: v20240317
Show newest version
/*
 * Copyright 2009 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.checkArgument;
import static com.google.common.base.Preconditions.checkState;

import com.google.common.base.Predicate;
import com.google.common.base.Predicates;
import com.google.javascript.jscomp.ControlFlowGraph.AbstractCfgNodeTraversalCallback;
import com.google.javascript.jscomp.ControlFlowGraph.Branch;
import com.google.javascript.jscomp.MustBeReachingVariableDef.Definition;
import com.google.javascript.jscomp.NodeTraversal.AbstractShallowCallback;
import com.google.javascript.jscomp.NodeTraversal.ScopedCallback;
import com.google.javascript.jscomp.graph.DiGraph.DiGraphEdge;
import com.google.javascript.jscomp.graph.DiGraph.DiGraphNode;
import com.google.javascript.rhino.Node;
import com.google.javascript.rhino.Token;
import java.util.Collection;
import java.util.HashSet;
import java.util.Iterator;
import java.util.LinkedHashSet;
import java.util.Set;

/**
 * Inline variables when possible. Using the information from
 * {@link MaybeReachingVariableUse} and {@link MustBeReachingVariableDef},
 * this pass attempts to inline a variable by placing the value at the
 * definition where the variable is used. The basic requirements for inlining
 * are the following:
 *
 * 
    *
  • There is exactly one reaching definition at the use of that variable *
  • *
  • There is exactly one use for that definition of the variable *
  • *
* *

Other requirements can be found in {@link Candidate#canInline}. Currently * this pass does not operate on the global scope due to compilation time. */ class FlowSensitiveInlineVariables implements CompilerPass, ScopedCallback { /** * Implementation: * * This pass first perform a traversal to gather a list of Candidates that * could be inlined using {@link GatherCandidates}. * * The second step involves verifying that each candidate is actually safe * to inline with {@link Candidate#canInline(Scope)} and finally perform * inlining using {@link Candidate#inlineVariable()}. * * The reason for the delayed evaluation of the candidates is because we * need two separate dataflow result. */ private final AbstractCompiler compiler; private final SideEffectPredicate sideEffectPredicate; // These two pieces of data is persistent in the whole execution of enter // scope. private ControlFlowGraph cfg; private Set candidates; private MustBeReachingVariableDef reachingDef; private MaybeReachingVariableUse reachingUses; private class SideEffectPredicate implements Predicate { // Check if there are side effects affecting the value of any of these names // (but not properties defined on that name) private final Set namesToCheck; public SideEffectPredicate() { namesToCheck = null; } public SideEffectPredicate(Set names) { this.namesToCheck = names; } @Override public boolean apply(Node n) { // When the node is null it means, we reached the implicit return // where the function returns (possibly without an return statement) if (n == null) { return false; } if (namesToCheck != null && n.isName() && namesToCheck.contains(n.getString()) && NodeUtil.isLValue(n)) { // the name is being written to. this is a problem, unless it is part of a top-level assign // chain and the write will take place after all CFG node subexpressions are evaluated return !isTopLevelAssignTarget(n); } AstAnalyzer astAnalyzer = compiler.getAstAnalyzer(); // TODO(user): We only care about calls to functions that // passes one of the dependent variable to a non-side-effect free // function. if (n.isCall() && astAnalyzer.functionCallHasSideEffects(n)) { return true; } if (n.isNew() && astAnalyzer.constructorCallHasSideEffects(n)) { return true; } if (n.isDelProp()) { return true; } for (Node c = n.getFirstChild(); c != null; c = c.getNext()) { if (!ControlFlowGraph.isEnteringNewCfgNode(c) && apply(c)) { return true; } } return false; } } /** Whether the given node is the target of a (possibly chained) assignment */ private static boolean isTopLevelAssignTarget(Node n) { Node ancestor = n.getParent(); while (ancestor.isAssign()) { ancestor = ancestor.getParent(); } return ancestor.isExprResult(); } public FlowSensitiveInlineVariables(AbstractCompiler compiler) { this.compiler = compiler; this.sideEffectPredicate = new SideEffectPredicate(); } @Override public final boolean shouldTraverse(NodeTraversal t, Node n, Node parent) { return !n.isScript() || !t.getInput().isExtern(); } @Override public void enterScope(NodeTraversal t) { if (t.inGlobalScope()) { return; // Don't even brother. All global variables are likely escaped. } if (!t.getScope().isFunctionBlockScope()) { return; // Only want to do the following if its a function block scope. } Node functionScopeRoot = t.getScopeRoot().getParent(); if (!isCandidateFunction(functionScopeRoot)) { return; } if (LiveVariablesAnalysis.MAX_VARIABLES_TO_ANALYZE < t.getScope().getVarCount()) { return; } SyntacticScopeCreator scopeCreator = (SyntacticScopeCreator) t.getScopeCreator(); // Compute the forward reaching definition. ControlFlowAnalysis cfa = new ControlFlowAnalysis(compiler, false, true); // Process the body of the function. cfa.process(null, functionScopeRoot); cfg = cfa.getCfg(); reachingDef = new MustBeReachingVariableDef(cfg, t.getScope(), compiler, scopeCreator); reachingDef.analyze(); candidates = new LinkedHashSet<>(); // Using the forward reaching definition search to find all the inline // candidates NodeTraversal.traverse(compiler, t.getScopeRoot(), new GatherCandidates()); // Compute the backward reaching use. The CFG can be reused. reachingUses = new MaybeReachingVariableUse(cfg, t.getScope(), compiler, scopeCreator); reachingUses.analyze(); while (!candidates.isEmpty()) { Candidate c = candidates.iterator().next(); if (c.canInline(t.getScope())) { c.inlineVariable(); candidates.remove(c); // If candidate "c" has dependencies, then inlining it may have introduced new dependencies // for our other inlining candidates. MustBeReachingVariableDef uses a dependency graph in // its analysis. Generating a new dependency graph will need another CFG computation. // Ideally we should iterate to a fixed point, but that can be costly. Therefore, we use // a conservative heuristic here: For each candidate "other", we back off if its set of // dependencies cannot contain all of "c"'s dependencies. if (!c.defMetadata.depends.isEmpty()) { for (Iterator it = candidates.iterator(); it.hasNext();) { Candidate other = it.next(); if (other.defMetadata.depends.contains(t.getScope().getVar(c.varName)) && !other.defMetadata.depends.containsAll(c.defMetadata.depends)) { it.remove(); } } } } else { candidates.remove(c); } } } private boolean isCandidateFunction(Node fn) { Node fnBody = fn.getLastChild(); return containsCandidateExpressions(fnBody); } private static boolean containsCandidateExpressions(Node n) { if (n.isFunction()) { // don't recurse into inner functions or into expressions the can't contain declarations. return false; } if (NodeUtil.isNameDeclaration(n) || isAssignmentToName(n)) { // if it is a simple assignment if (n.getFirstChild().isName()) { return true; } } for (Node c = n.getFirstChild(); c != null; c = c.getNext()) { if (containsCandidateExpressions(c)) { return true; } } return false; } private static boolean isAssignmentToName(Node n) { if (NodeUtil.isAssignmentOp(n) || n.isDec() || n.isInc()) { // if it is a simple assignment return (n.getFirstChild().isName()); } return false; } @Override public void exitScope(NodeTraversal t) {} @Override public void process(Node externs, Node root) { (new NodeTraversal(compiler, this, new SyntacticScopeCreator(compiler))) .traverseRoots(externs, root); } @Override public void visit(NodeTraversal t, Node n, Node parent) { // TODO(user): While the helpers do a subtree traversal on the AST, the // compiler pass itself only traverse the AST to look for function // declarations to perform dataflow analysis on. We could combine // the traversal in DataFlowAnalysis's computeEscaped later to save some // time. } private class GatherCandidatesCfgNodeCallback extends AbstractCfgNodeTraversalCallback { Node cfgNode = null; public void setCfgNode(Node cfgNode) { this.cfgNode = cfgNode; } @Override public void visit(NodeTraversal t, Node n, Node parent) { if (n.isName()) { // n.getParent() isn't null. This just the case where n is the root // node that gatherCb started at. if (parent == null) { return; } // Make sure that the name node is purely a read. if ((NodeUtil.isAssignmentOp(parent) && parent.getFirstChild() == n) || NodeUtil.isNameDeclaration(parent) || parent.isInc() || parent.isDec() || parent.isParamList() || parent.isCatch() || NodeUtil.isLhsByDestructuring(n)) { return; } String name = n.getString(); if (compiler.getCodingConvention().isExported(name)) { return; } Definition def = reachingDef.getDef(name, cfgNode); // TODO(nicksantos): We need to add some notion of @const outer // scope vars. We can inline those just fine. if (def != null && !reachingDef.dependsOnOuterScopeVars(def)) { candidates.add(new Candidate(name, def, n, cfgNode)); } } } } /** * Gathers a list of possible candidates for inlining based only on * information from {@link MustBeReachingVariableDef}. The list will be stored * in {@code candidates} and the validity of each inlining Candidate should * be later verified with {@link Candidate#canInline(Scope)} when * {@link MaybeReachingVariableUse} has been performed. */ private class GatherCandidates extends AbstractShallowCallback { final GatherCandidatesCfgNodeCallback gatherCb = new GatherCandidatesCfgNodeCallback(); @Override public void visit(NodeTraversal t, Node n, Node parent) { DiGraphNode graphNode = cfg.getNode(n); if (graphNode == null) { // Not a CFG node. return; } final Node cfgNode = n; gatherCb.setCfgNode(cfgNode); NodeTraversal.traverse(compiler, cfgNode, gatherCb); } } /** * Models the connection between a definition and a use of that definition. */ private class Candidate { // Name of the variable. private final String varName; // Nodes related to the definition. private Node def; private final Definition defMetadata; // Nodes related to the use. private final Node use; private final Node useCfgNode; // Number of uses of the variable within the current CFG node. private int numUsesWithinCfgNode; Candidate(String varName, Definition defMetadata, Node use, Node useCfgNode) { checkArgument(use.isName()); this.varName = varName; this.defMetadata = defMetadata; this.use = use; this.useCfgNode = useCfgNode; } private Node getDefCfgNode() { return defMetadata.node; } private boolean canInline(final Scope scope) { // Cannot inline a parameter. if (getDefCfgNode().isFunction()) { return false; } getDefinition(getDefCfgNode()); getNumUseInUseCfgNode(useCfgNode); // Definition was not found. if (def == null) { return false; } // Check that the assignment isn't used as a R-Value. // TODO(user): Certain cases we can still inline. if (def.isAssign() && !NodeUtil.isExprAssign(def.getParent())) { return false; } Set namesToCheck = new HashSet<>(); if (defMetadata.depends != null) { for (Var var : defMetadata.depends) { namesToCheck.add(var.getName()); } } SideEffectPredicate sideEffectPredicateWithNames = new SideEffectPredicate(namesToCheck); // A subexpression evaluated after the variable has a side effect. // Example, for x: // x = readProp(b), modifyProp(b); print(x); if (checkPostExpressions(def, getDefCfgNode(), sideEffectPredicateWithNames)) { return false; } // Similar check as the above but this time, all the sub-expressions // evaluated before the variable. // x = readProp(b); modifyProp(b), print(x); if (checkPreExpressions(use, useCfgNode, sideEffectPredicateWithNames)) { return false; } // TODO(user): Side-effect is OK sometimes. As long as there are no // side-effect function down all paths to the use. Once we have all the // side-effect analysis tool. if (compiler.getAstAnalyzer().mayHaveSideEffects(def.getLastChild())) { return false; } // TODO(user): We could inline all the uses if the expression is short. // Finally we have to make sure that there are no more than one use // in the program and in the CFG node. Even when it is semantically // correctly inlining twice increases code size. if (numUsesWithinCfgNode != 1) { return false; } // Make sure that the name is not within a loop if (NodeUtil.isWithinLoop(use)) { return false; } Collection uses = reachingUses.getUses(varName, getDefCfgNode()); if (uses.size() != 1) { return false; } if (!isRhsSafeToInline(scope)) { return false; } // We can skip the side effect check along the paths of two nodes if // they are just next to each other. if (NodeUtil.isStatementBlock(getDefCfgNode().getParent()) && getDefCfgNode().getNext() != useCfgNode) { // Similar side effect check as above but this time the side effect is // else where along the path. // x = readProp(b); while(modifyProp(b)) {}; print(x); CheckPathsBetweenNodes pathCheck = new CheckPathsBetweenNodes<>( cfg, cfg.getNode(getDefCfgNode()), cfg.getNode(useCfgNode), sideEffectPredicate, Predicates.>alwaysTrue(), false); if (pathCheck.somePathsSatisfyPredicate()) { return false; } } return true; } /** * Actual transformation. */ private void inlineVariable() { Node defParent = def.getParent(); Node useParent = use.getParent(); if (def.isAssign()) { Node rhs = def.getLastChild(); rhs.detach(); // Oh yes! I have grandparent to remove this. checkState(defParent.isExprResult()); while (defParent.getParent().isLabel()) { defParent = defParent.getParent(); } compiler.reportChangeToEnclosingScope(defParent); defParent.detach(); useParent.replaceChild(use, rhs); } else if (NodeUtil.isNameDeclaration(defParent)) { Node rhs = def.getLastChild(); if (defParent.isConst()) { // If it is a const var we don't want to remove the rhs of the variable def.replaceChild(rhs, Node.newString(Token.NAME, "undefined")); useParent.replaceChild(use, rhs); } else { def.removeChild(rhs); useParent.replaceChild(use, rhs); } } else { throw new IllegalStateException("No other definitions can be inlined."); } compiler.reportChangeToEnclosingScope(useParent); } /** * Set the def node * * @param n A node that has a corresponding CFG node in the CFG. */ private void getDefinition(Node n) { AbstractCfgNodeTraversalCallback gatherCb = new AbstractCfgNodeTraversalCallback() { @Override public void visit(NodeTraversal t, Node n, Node parent) { switch (n.getToken()) { case NAME: if (n.getString().equals(varName) && n.hasChildren()) { def = n; } return; case ASSIGN: Node lhs = n.getFirstChild(); if (lhs.isName() && lhs.getString().equals(varName)) { def = n; } return; default: break; } } }; NodeTraversal.traverse(compiler, n, gatherCb); } /** * Computes the number of uses of the variable varName and store it in * numUseWithinUseCfgNode. */ private void getNumUseInUseCfgNode(final Node cfgNode) { numUsesWithinCfgNode = 0; AbstractCfgNodeTraversalCallback gatherCb = new AbstractCfgNodeTraversalCallback() { @Override public void visit(NodeTraversal t, Node n, Node parent) { if (n.isName() && n.getString().equals(varName)) { // We make a special exception when the entire cfgNode is a chain // of assignments, since in that case the assignment statements // will happen after the inlining of the right hand side. // TODO(lharker): We can probably remove the isAssignChain check, and instead use // the SideEffectPredicate to look for dangerous assignments in the same CFG node if (parent.isAssign() && (parent.getFirstChild() == n) && isAssignChain(parent, cfgNode)) { // Don't count lhs of top-level assignment chain return; } else { numUsesWithinCfgNode++; } } } private boolean isAssignChain(Node child, Node ancestor) { for (Node n = child; n != ancestor; n = n.getParent()) { if (!n.isAssign()) { return false; } } return true; } }; NodeTraversal.traverse(compiler, cfgNode, gatherCb); } /** * Check if the definition we're considering inline has anything that makes inlining unsafe * (that hasn't already been caught). * * @param usageScope The scope we will inline the variable into. */ private boolean isRhsSafeToInline(final Scope usageScope) { // Don't inline definitions with an R-Value that has: // 1) GETPROP, GETELEM, // 2) anything that creates a new object. // Example: // var x = a.b.c; j.c = 1; print(x); // Inlining print(a.b.c) is not safe - consider if j were an alias to a.b. if (NodeUtil.has( def.getLastChild(), new Predicate() { @Override public boolean apply(Node input) { switch (input.getToken()) { case GETELEM: case GETPROP: case ARRAYLIT: case OBJECTLIT: case REGEXP: case NEW: return true; // unsafe to inline. default: break; } return false; } }, new Predicate() { @Override public boolean apply(Node input) { // Recurse if the node is not a function. return !input.isFunction(); } })) { return false; } // Don't inline definitions with an rvalue referencing names that are not declared in the // usage's scope. (Unlike the above check, this includes names referenced inside function // expressions in the rvalue). // e.g. the name "a" below in the definition of "b": // { // let a = 3; // var b = a; // } // return b; // "a" is not declared in this scope so we can't inline this to "return a;" if (NodeUtil.has( def.getLastChild(), new Predicate() { @Override public boolean apply(Node input) { if (input.isName()) { String name = input.getString(); if (!name.isEmpty() && !usageScope.hasSlot(name)) { return true; // unsafe to inline. } } return false; } }, Predicates.alwaysTrue())) { return false; } return true; } } /** * Given an expression by its root and sub-expression n, return true if the predicate is true for * some expression evaluated after n. * *

NOTE: this doesn't correctly check destructuring patterns, because their order of evaluation * is different from AST traversal order, but currently this is ok because * FlowSensitiveInlineVariables never inlines variable assignments inside destructuring. * *

Example: * *

NotChecked(), NotChecked(), n, Checked(), Checked(); */ private static boolean checkPostExpressions( Node n, Node expressionRoot, Predicate predicate) { for (Node p = n; p != expressionRoot; p = p.getParent()) { for (Node cur = p.getNext(); cur != null; cur = cur.getNext()) { if (predicate.apply(cur)) { return true; } } } return false; } /** * Given an expression by its root and sub-expression n, return true if the predicate is true for * some expression evaluated before n. * *

In most cases evaluation order follows left-to-right AST order. Destructuring pattern * evaluation is an exception. * *

Example: * *

Checked(), Checked(), n, NotChecked(), NotChecked(); */ private static boolean checkPreExpressions( Node n, Node expressionRoot, Predicate predicate) { for (Node p = n; p != expressionRoot; p = p.getParent()) { Node oldestSibling = p.getParent().getFirstChild(); // Evaluate a destructuring assignment right-to-left. if (oldestSibling.isDestructuringPattern()) { if (p.isDestructuringPattern()) { if (p.getNext() != null && predicate.apply(p.getNext())) { return true; } } continue; } for (Node cur = oldestSibling; cur != p; cur = cur.getNext()) { if (predicate.apply(cur)) { return true; } } } return false; } }





© 2015 - 2024 Weber Informatics LLC | Privacy Policy