com.google.javascript.jscomp.GuardedCallback Maven / Gradle / Ivy
Show all versions of closure-compiler Show documentation
/*
* Copyright 2016 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 com.google.common.collect.HashMultiset;
import com.google.common.collect.ImmutableSet;
import com.google.common.collect.ListMultimap;
import com.google.common.collect.MultimapBuilder;
import com.google.common.collect.Multiset;
import com.google.common.collect.SetMultimap;
import com.google.common.collect.Sets;
import com.google.javascript.jscomp.NodeTraversal.Callback;
import com.google.javascript.rhino.Node;
import com.google.javascript.rhino.Token;
import java.util.ArrayDeque;
import java.util.Deque;
/**
* An AST traverser that keeps track of whether access to a generic resource are "guarded" or not. A
* guarded use is one that will not cause runtime errors if the resource does not exist. The
* resource (type {@code T} in the signature) is any value type computable from a node, such as a
* qualified name or property name. It is up to the subclass to map the currently visited token to
* the {@code T} in order to call {@link #isGuarded}, which depends entirely on the context from
* ancestor nodes and previous calls to {@code #isGuarded} for the same resource.
*
* More precisely, a resource may be guarded either intrinsically or conditionally,
* as follows. A use is intrinsically guarded if it occurs in a context where its specific value is
* immediately discarded, such as coercion to boolean or string. A use is conditionally guarded if
* it occurs in a context conditioned on an intrinsically guarded use (such as the "then" or "else"
* block of an "if", the second or third argument of a ternary operator, right-hand arguments of
* logical operators, or later in a block in which an unconditional "throw" or "return" was found in
* a guarded context).
*
*
For example, the following are all intrinsically guarded uses of {@code x}:
*
*
{@code
* // Coerced to boolean:
* if (x);
* x ? y : z;
* x && y;
* Boolean(x);
* !x;
* x == y; x != y; x === y; x !== y;
* x instanceof Foo;
* typeof x === 'string';
*
* // Coerced to string:
* String(x);
* typeof x;
*
* // Immediately discarded (but doesn't make much sense in my contexts):
* x = y;
* }
*
* The following uses of {@code x} are conditionally guarded:
*
*
{@code
* if (x) x();
* !x ? null : x;
* typeof x == 'function' ? x : () => {};
* x && x.y;
* !x || x.y;
* if (!x) return; x();
* x ?? x = 3;
* }
*
* Note that there is no logic to determine which branch is guarded, so any usages in either
* branch will pass after such a check. As such, the following are also considered guarded, though a
* human can easily see that this is spurious:
*
* {@code
* if (!x) x();
* if (x) { } else x();
* !x && x();
* var y = x != null ? null : x;
* }
*
* Note also that the call or property access is not necessary to make a use unguarded: the final
* example immediately above would be unguarded if it weren't for the {@code x != null} condition,
* since it allows the value of {@code x} to leak out in an uncontrolled way.
*
* This class overrides the {@link Callback} API methods with final methods of its own, and
* defines the template method {@link #visitGuarded} to perform the normal work for individual
* nodes. The only other API is {@link #isGuarded}, which allows checking if a {@code T} in the
* current node's context is guarded, either intrinsically or conditionally. If it is intrinsically
* guarded, then it may be recorded as a condition for the purpose of guarding future contexts.
*/
abstract class GuardedCallback implements Callback {
// Compiler is needed for coding convention (isPropertyTestFunction).
private final AbstractCompiler compiler;
// Map from short-circuiting conditional nodes (AND, OR, COALESCE, IF, and HOOK) to
// the set of resources each node guards. This is saved separately from
// just `guarded` because the guard must not go into effect until after
// traversal of the first child is complete. Before traversing the second
// child any node, its values in this map are moved into `guarded` and
// `installedGuards` (the latter allowing removal at the correct time).
private final SetMultimap registeredGuards =
MultimapBuilder.hashKeys().hashSetValues().build();
// Set of currently-guarded resources. Elements are added to this set
// just before traversing the second or later (i.e. "then" or "else")
// child of a short-circuiting conditional node, and then removed after
// traversing the last child. It is a multiset so that multiple adds
// of the same resource require the same number of removals before the
// resource becomes unguarded.
private final Multiset guarded = HashMultiset.create();
// Resources that are currently installed as guarded but will need to
// be removed from `guarded` after visiting all the key nodes' children.
private final ListMultimap installedGuards =
MultimapBuilder.hashKeys().arrayListValues().build();
// A stack of `Context` objects describing the current node's context:
// specifically, whether it is inherently safe, and a link to one or
// more conditional nodes in the current statement directly above it
// (for registering safe resources as guards).
private final Deque contextStack = new ArrayDeque<>();
GuardedCallback(AbstractCompiler compiler) {
this.compiler = compiler;
}
@Override
public final boolean shouldTraverse(NodeTraversal traversal, Node n, Node parent) {
// Note that shouldTraverse() operates primarily on `parent`, while visit()
// uses `n`. This is intentional. To see why, consider traversing the
// following tree:
//
// if (x) y; else z;
//
// 1. shouldTraverse(`if`):
// a. parent is null, so pushes an EMPTY onto the context stack.
// 2. shouldTraverse(`x`):
// a. parent is `if`, so pushes Context(`if`, true); guards is empty.
// 3. visit(`x`)
// a. guarded and installedGuards are both empty, so nothing is removed.
// b. visitGuarded(`x`) will call isGuarded("x"), which looks at the top
// of the stack and sees that the context is safe (true) and that
// there is a linked conditional node (the `if`); adds {`if`: "x"}
// to registeredGuards.
// b. Context(`if`, true) is popped off the stack.
// 4. shouldTraverse(`y`):
// a. parent is still `if`, but since `y` is the second child it is
// no longer safe, so another EMPTY is pushed.
// b. the {`if`: "x"} guard is moved from registered to installed.
// 5. visit(`y`):
// a. nothing is installed on `y` so no guards are removed.
// b. visitGuarded(`y`) will call isGuarded("y"), which will return
// false since "y" is neither intrinsically or conditionally guarded;
// if we'd called isResourceRequired("x"), it would return false
// because "x" is currently an element of guarded.
// c. one empty context is popped.
// 6. shouldTraverse(`z`), visit(`z`)
// a. see steps 4-5, nothing really changes here.
// 7. visit(`if`)
// a. the installed {`if`: "x"} guard is removed.
// c. pop the final empty context from the stack.
if (parent == null) {
// The first node gets an empty context.
contextStack.push(Context.EMPTY);
} else {
// Before traversing any children, we update the stack
contextStack.push(contextStack.peek().descend(compiler, parent, n));
// If the parent has any guards registered on it, then add them to both
// `guarded` and `installedGuards`.
if (parent != null && CAN_HAVE_GUARDS.contains(parent.getToken())
&& registeredGuards.containsKey(parent)) {
for (T resource : registeredGuards.removeAll(parent)) {
guarded.add(resource);
installedGuards.put(parent, resource);
}
}
}
return true;
}
@Override
public final void visit(NodeTraversal traversal, Node n, Node parent) {
// Remove any guards registered on this node by its children, which are no longer
// relevant. This happens first because these were registered on a "parent", but
// now this is that parent (i.e. `n` here vs `parent` in isGuarded).
if (parent != null && CAN_HAVE_GUARDS.contains(n.getToken())
&& installedGuards.containsKey(n)) {
guarded.removeAll(installedGuards.removeAll(n));
}
// Check for abrupt returns (`return` and `throw`).
if (isAbrupt(n)) {
// If found, any guards installed on a parent IF should be promoted to the
// grandparent. This allows a small amount of flow-sensitivity, in that
// if (!x) return; x();
// has the guard for `x` promoted from the `if` to the outer block, so that
// it guards the next statement.
promoteAbruptReturns(parent);
}
// Finally continue on to whatever the traversal would normally do.
visitGuarded(traversal, n, parent);
// After children have been traversed, pop the top of the conditional stack.
contextStack.pop();
}
private void promoteAbruptReturns(Node parent) {
// If the parent is a BLOCK (e.g. `if (x) { return; }`) then go up one level.
if (parent.isBlock()) {
parent = parent.getParent();
}
// If there were any guards registered the parent IF, then promote them up one level.
if (parent.isIf() && installedGuards.containsKey(parent)) {
Node grandparent = parent.getParent();
if (grandparent.isBlock() || grandparent.isScript()) {
registeredGuards.putAll(grandparent, installedGuards.get(parent));
}
}
}
/**
* Performs specific traversal behavior. Should call {@link #isGuarded}
* at least once.
*/
abstract void visitGuarded(NodeTraversal traversal, Node n, Node parent);
/**
* Determines if the given resource is guarded, either intrinsically or
* conditionally. If the former, any ancestor conditional nodes are
* registered as feature-testing the resource.
*/
boolean isGuarded(T resource) {
// Check if this polyfill is already guarded. If so, return true right away.
if (guarded.contains(resource)) {
return true;
}
// If not, see if this is itself a feature check guard. This is
// defined as a usage of the polyfill in such a way that throws
// away the actual value and only cares about its truthiness or
// typeof. We walk up the ancestor tree through a small set of
// node types and if this is detected to be a guard, then the
// conditional node is marked as a guard for this polyfill.
Context context = contextStack.peek();
if (!context.safe) {
return false;
}
// Loop over all the linked conditionals and register this as a guard.
while (context != null && context.conditional != null) {
registeredGuards.put(context.conditional, resource);
context = context.linked;
}
return true;
}
// The context of a node, keeping track of whether it is safe for
// possibly-undefined values, and whether there are any conditionals
// upstream in the tree.
private static class Context {
// An empty instance: unsafe and with no linked conditional nodes.
static final Context EMPTY = new Context(null, false, null);
// The most recent conditional.
final Node conditional;
// Whether this position is safe for an undefined type.
final boolean safe;
// A very naive linked list for storing additional conditional nodes.
final Context linked;
Context(Node conditional, boolean safe, Context linked) {
this.conditional = conditional;
this.safe = safe;
this.linked = linked;
}
// Returns a new Context with a new conditional node and safety status.
// If the current context already has a conditional, then it is linked
// so that both can be marked when necessary.
Context link(Node newConditional, boolean newSafe) {
return new Context(newConditional, newSafe, this.conditional != null ? this : null);
}
// Returns a new Context with a different safety bit, but doesn't
// change anything else.
Context propagate(boolean newSafe) {
return newSafe == safe ? this : new Context(conditional, newSafe, linked);
}
// Returns a new context given the current context and the next parent
// node. Child is only used to determine whether we're looking at the
// first child or not.
Context descend(AbstractCompiler compiler, Node parent, Node child) {
boolean first = child == parent.getFirstChild();
switch (parent.getToken()) {
case CAST:
// Casts are irrelevant.
return this;
case COMMA:
// `Promise, whatever` is safe.
// `whatever, Promise` is same as outer context.
return child == parent.getLastChild() ? this : propagate(true);
case AND:
// `Promise && whatever` never returns Promise itself, so it is safe.
// `whatever && Promise` may return Promise, so return outer context.
return first ? link(parent, true) : this;
case OR:
case COALESCE:
// throw new RuntimeException("beeeeep");
// `Promise || whatever` and `Promise ?? whatever`
// may return Promise (unsafe), but is itself a conditional.
// `whatever || Promise` and `whatever ?? Promise`
// is same as outer context.
return first ? link(parent, false) : this;
case HOOK:
// `Promise ? whatever : whatever` is a safe conditional.
// `whatever ? Promise : whatever` (etc) is same as outer context.
return first ? link(parent, true) : this;
case IF:
// `if (Promise) whatever` is a safe conditional.
// `if (whatever) { ... }` is nothing.
// TODO(sdh): Handle do/while/for/for-of/for-in?
return first ? link(parent, true) : EMPTY;
case INSTANCEOF:
case ASSIGN:
// `Promise instanceof whatever` is safe, `whatever instanceof Promise` is not.
// `Promise = whatever` is a bad idea, but it's safe w.r.t. polyfills.
return propagate(first);
case TYPEOF:
case NOT:
case EQ:
case NE:
case SHEQ:
case SHNE:
// `typeof Promise` is always safe, as is `Promise == whatever`, etc.
return propagate(true);
case CALL:
// `String(Promise)` is safe, `Promise(whatever)` or `whatever(Promise)` is not.
return propagate(!first && isPropertyTestFunction(compiler, parent));
case ROOT:
// This case causes problems for isStatement() so handle it separately.
return EMPTY;
default:
// Expressions propagate linked conditionals; statements do not.
return NodeUtil.isStatement(parent) ? EMPTY : propagate(false);
}
}
}
private static boolean isAbrupt(Node n) {
return n.isReturn() || n.isThrow();
}
// Extend the coding convention's idea of property test functions to also
// include String() and Boolean().
private static boolean isPropertyTestFunction(AbstractCompiler compiler, Node n) {
if (compiler.getCodingConvention().isPropertyTestFunction(n)) {
return true;
}
Node target = n.getFirstChild();
return target.isName() && PROPERTY_TEST_FUNCTIONS.contains(target.getString());
}
// NOTE: we currently assume these are simple (unqualified) names.
private static final ImmutableSet PROPERTY_TEST_FUNCTIONS =
ImmutableSet.of("String", "Boolean");
// Tokens that are allowed to have guards on them (no point doing a hash lookup on
// any other type of node).
private static final ImmutableSet CAN_HAVE_GUARDS =
Sets.immutableEnumSet(
Token.AND, Token.OR, Token.COALESCE, Token.HOOK, Token.IF, Token.BLOCK, Token.SCRIPT);
}