forked from mirrors/gecko-dev
1836 lines
59 KiB
C++
1836 lines
59 KiB
C++
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
|
|
* vim: set ts=8 sts=4 et sw=4 tw=99:
|
|
* This Source Code Form is subject to the terms of the Mozilla Public
|
|
* License, v. 2.0. If a copy of the MPL was not distributed with this
|
|
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
|
|
|
|
#include "frontend/FoldConstants.h"
|
|
|
|
#include "mozilla/FloatingPoint.h"
|
|
|
|
#include "jslibmath.h"
|
|
#include "jsnum.h"
|
|
|
|
#include "frontend/ParseNode.h"
|
|
#include "frontend/Parser.h"
|
|
#include "js/Conversions.h"
|
|
#include "vm/StringType.h"
|
|
|
|
using namespace js;
|
|
using namespace js::frontend;
|
|
|
|
using mozilla::IsNaN;
|
|
using mozilla::IsNegative;
|
|
using mozilla::NegativeInfinity;
|
|
using mozilla::PositiveInfinity;
|
|
using JS::GenericNaN;
|
|
using JS::ToInt32;
|
|
using JS::ToUint32;
|
|
|
|
static bool
|
|
ContainsHoistedDeclaration(JSContext* cx, ParseNode* node, bool* result);
|
|
|
|
static bool
|
|
ListContainsHoistedDeclaration(JSContext* cx, ListNode* list, bool* result)
|
|
{
|
|
for (ParseNode* node = list->pn_head; node; node = node->pn_next) {
|
|
if (!ContainsHoistedDeclaration(cx, node, result))
|
|
return false;
|
|
if (*result)
|
|
return true;
|
|
}
|
|
|
|
*result = false;
|
|
return true;
|
|
}
|
|
|
|
// Determines whether the given ParseNode contains any declarations whose
|
|
// visibility will extend outside the node itself -- that is, whether the
|
|
// ParseNode contains any var statements.
|
|
//
|
|
// THIS IS NOT A GENERAL-PURPOSE FUNCTION. It is only written to work in the
|
|
// specific context of deciding that |node|, as one arm of a ParseNodeKind::If
|
|
// controlled by a constant condition, contains a declaration that forbids
|
|
// |node| being completely eliminated as dead.
|
|
static bool
|
|
ContainsHoistedDeclaration(JSContext* cx, ParseNode* node, bool* result)
|
|
{
|
|
if (!CheckRecursionLimit(cx))
|
|
return false;
|
|
|
|
restart:
|
|
|
|
// With a better-typed AST, we would have distinct parse node classes for
|
|
// expressions and for statements and would characterize expressions with
|
|
// ExpressionKind and statements with StatementKind. Perhaps someday. In
|
|
// the meantime we must characterize every ParseNodeKind, even the
|
|
// expression/sub-expression ones that, if we handle all statement kinds
|
|
// correctly, we'll never see.
|
|
switch (node->getKind()) {
|
|
// Base case.
|
|
case ParseNodeKind::Var:
|
|
*result = true;
|
|
return true;
|
|
|
|
// Non-global lexical declarations are block-scoped (ergo not hoistable).
|
|
case ParseNodeKind::Let:
|
|
case ParseNodeKind::Const:
|
|
MOZ_ASSERT(node->isArity(PN_LIST));
|
|
*result = false;
|
|
return true;
|
|
|
|
// Similarly to the lexical declarations above, classes cannot add hoisted
|
|
// declarations
|
|
case ParseNodeKind::Class:
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
*result = false;
|
|
return true;
|
|
|
|
// Function declarations *can* be hoisted declarations. But in the
|
|
// magical world of the rewritten frontend, the declaration necessitated
|
|
// by a nested function statement, not at body level, doesn't require
|
|
// that we preserve an unreachable function declaration node against
|
|
// dead-code removal.
|
|
case ParseNodeKind::Function:
|
|
MOZ_ASSERT(node->isArity(PN_CODE));
|
|
*result = false;
|
|
return true;
|
|
|
|
case ParseNodeKind::Module:
|
|
*result = false;
|
|
return true;
|
|
|
|
// Statements with no sub-components at all.
|
|
case ParseNodeKind::EmptyStatement:
|
|
case ParseNodeKind::Debugger:
|
|
MOZ_ASSERT(node->isArity(PN_NULLARY));
|
|
*result = false;
|
|
return true;
|
|
|
|
// Statements containing only an expression have no declarations.
|
|
case ParseNodeKind::ExpressionStatement:
|
|
case ParseNodeKind::Throw:
|
|
case ParseNodeKind::Return:
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
*result = false;
|
|
return true;
|
|
|
|
// These two aren't statements in the spec, but we sometimes insert them
|
|
// in statement lists anyway.
|
|
case ParseNodeKind::InitialYield:
|
|
case ParseNodeKind::YieldStar:
|
|
case ParseNodeKind::Yield:
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
*result = false;
|
|
return true;
|
|
|
|
// Other statements with no sub-statement components.
|
|
case ParseNodeKind::Break:
|
|
case ParseNodeKind::Continue:
|
|
case ParseNodeKind::Import:
|
|
case ParseNodeKind::ImportSpecList:
|
|
case ParseNodeKind::ImportSpec:
|
|
case ParseNodeKind::ExportFrom:
|
|
case ParseNodeKind::ExportDefault:
|
|
case ParseNodeKind::ExportSpecList:
|
|
case ParseNodeKind::ExportSpec:
|
|
case ParseNodeKind::Export:
|
|
case ParseNodeKind::ExportBatchSpec:
|
|
*result = false;
|
|
return true;
|
|
|
|
// Statements possibly containing hoistable declarations only in the left
|
|
// half, in ParseNode terms -- the loop body in AST terms.
|
|
case ParseNodeKind::DoWhile:
|
|
return ContainsHoistedDeclaration(cx, node->pn_left, result);
|
|
|
|
// Statements possibly containing hoistable declarations only in the
|
|
// right half, in ParseNode terms -- the loop body or nested statement
|
|
// (usually a block statement), in AST terms.
|
|
case ParseNodeKind::While:
|
|
case ParseNodeKind::With:
|
|
return ContainsHoistedDeclaration(cx, node->pn_right, result);
|
|
|
|
case ParseNodeKind::Label:
|
|
return ContainsHoistedDeclaration(cx, node->pn_expr, result);
|
|
|
|
// Statements with more complicated structures.
|
|
|
|
// if-statement nodes may have hoisted declarations in their consequent
|
|
// and alternative components.
|
|
case ParseNodeKind::If: {
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
|
|
ParseNode* consequent = node->pn_kid2;
|
|
if (!ContainsHoistedDeclaration(cx, consequent, result))
|
|
return false;
|
|
if (*result)
|
|
return true;
|
|
|
|
if ((node = node->pn_kid3))
|
|
goto restart;
|
|
|
|
*result = false;
|
|
return true;
|
|
}
|
|
|
|
// try-statements have statements to execute, and one or both of a
|
|
// catch-list and a finally-block.
|
|
case ParseNodeKind::Try: {
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
MOZ_ASSERT(node->pn_kid2 || node->pn_kid3,
|
|
"must have either catch(es) or finally");
|
|
|
|
ParseNode* tryBlock = node->pn_kid1;
|
|
if (!ContainsHoistedDeclaration(cx, tryBlock, result))
|
|
return false;
|
|
if (*result)
|
|
return true;
|
|
|
|
if (ParseNode* catchScope = node->pn_kid2) {
|
|
MOZ_ASSERT(catchScope->isKind(ParseNodeKind::LexicalScope));
|
|
|
|
ParseNode* catchNode = catchScope->pn_expr;
|
|
MOZ_ASSERT(catchNode->isKind(ParseNodeKind::Catch));
|
|
|
|
ParseNode* catchStatements = catchNode->pn_right;
|
|
if (!ContainsHoistedDeclaration(cx, catchStatements, result))
|
|
return false;
|
|
if (*result)
|
|
return true;
|
|
}
|
|
|
|
if (ParseNode* finallyBlock = node->pn_kid3)
|
|
return ContainsHoistedDeclaration(cx, finallyBlock, result);
|
|
|
|
*result = false;
|
|
return true;
|
|
}
|
|
|
|
// A switch node's left half is an expression; only its right half (a
|
|
// list of cases/defaults, or a block node) could contain hoisted
|
|
// declarations.
|
|
case ParseNodeKind::Switch:
|
|
MOZ_ASSERT(node->isArity(PN_BINARY));
|
|
return ContainsHoistedDeclaration(cx, node->pn_right, result);
|
|
|
|
case ParseNodeKind::Case:
|
|
return ContainsHoistedDeclaration(cx, node->as<CaseClause>().statementList(), result);
|
|
|
|
case ParseNodeKind::For: {
|
|
MOZ_ASSERT(node->isArity(PN_BINARY));
|
|
|
|
ParseNode* loopHead = node->pn_left;
|
|
MOZ_ASSERT(loopHead->isKind(ParseNodeKind::ForHead) ||
|
|
loopHead->isKind(ParseNodeKind::ForIn) ||
|
|
loopHead->isKind(ParseNodeKind::ForOf));
|
|
|
|
if (loopHead->isKind(ParseNodeKind::ForHead)) {
|
|
// for (init?; cond?; update?), with only init possibly containing
|
|
// a hoisted declaration. (Note: a lexical-declaration |init| is
|
|
// (at present) hoisted in SpiderMonkey parlance -- but such
|
|
// hoisting doesn't extend outside of this statement, so it is not
|
|
// hoisting in the sense meant by ContainsHoistedDeclaration.)
|
|
MOZ_ASSERT(loopHead->isArity(PN_TERNARY));
|
|
|
|
ParseNode* init = loopHead->pn_kid1;
|
|
if (init && init->isKind(ParseNodeKind::Var)) {
|
|
*result = true;
|
|
return true;
|
|
}
|
|
} else {
|
|
MOZ_ASSERT(loopHead->isKind(ParseNodeKind::ForIn) ||
|
|
loopHead->isKind(ParseNodeKind::ForOf));
|
|
|
|
// for each? (target in ...), where only target may introduce
|
|
// hoisted declarations.
|
|
//
|
|
// -- or --
|
|
//
|
|
// for (target of ...), where only target may introduce hoisted
|
|
// declarations.
|
|
//
|
|
// Either way, if |target| contains a declaration, it's |loopHead|'s
|
|
// first kid.
|
|
MOZ_ASSERT(loopHead->isArity(PN_TERNARY));
|
|
|
|
ParseNode* decl = loopHead->pn_kid1;
|
|
if (decl && decl->isKind(ParseNodeKind::Var)) {
|
|
*result = true;
|
|
return true;
|
|
}
|
|
}
|
|
|
|
ParseNode* loopBody = node->pn_right;
|
|
return ContainsHoistedDeclaration(cx, loopBody, result);
|
|
}
|
|
|
|
case ParseNodeKind::LexicalScope: {
|
|
MOZ_ASSERT(node->isArity(PN_SCOPE));
|
|
ParseNode* expr = node->pn_expr;
|
|
|
|
if (expr->isKind(ParseNodeKind::For) || expr->isKind(ParseNodeKind::Function))
|
|
return ContainsHoistedDeclaration(cx, expr, result);
|
|
|
|
MOZ_ASSERT(expr->isKind(ParseNodeKind::StatementList));
|
|
return ListContainsHoistedDeclaration(cx, &node->pn_expr->as<ListNode>(), result);
|
|
}
|
|
|
|
// List nodes with all non-null children.
|
|
case ParseNodeKind::StatementList:
|
|
return ListContainsHoistedDeclaration(cx, &node->as<ListNode>(), result);
|
|
|
|
// Grammar sub-components that should never be reached directly by this
|
|
// method, because some parent component should have asserted itself.
|
|
case ParseNodeKind::ObjectPropertyName:
|
|
case ParseNodeKind::ComputedName:
|
|
case ParseNodeKind::Spread:
|
|
case ParseNodeKind::MutateProto:
|
|
case ParseNodeKind::Colon:
|
|
case ParseNodeKind::Shorthand:
|
|
case ParseNodeKind::Conditional:
|
|
case ParseNodeKind::TypeOfName:
|
|
case ParseNodeKind::TypeOfExpr:
|
|
case ParseNodeKind::Await:
|
|
case ParseNodeKind::Void:
|
|
case ParseNodeKind::Not:
|
|
case ParseNodeKind::BitNot:
|
|
case ParseNodeKind::DeleteName:
|
|
case ParseNodeKind::DeleteProp:
|
|
case ParseNodeKind::DeleteElem:
|
|
case ParseNodeKind::DeleteExpr:
|
|
case ParseNodeKind::Pos:
|
|
case ParseNodeKind::Neg:
|
|
case ParseNodeKind::PreIncrement:
|
|
case ParseNodeKind::PostIncrement:
|
|
case ParseNodeKind::PreDecrement:
|
|
case ParseNodeKind::PostDecrement:
|
|
case ParseNodeKind::Or:
|
|
case ParseNodeKind::And:
|
|
case ParseNodeKind::BitOr:
|
|
case ParseNodeKind::BitXor:
|
|
case ParseNodeKind::BitAnd:
|
|
case ParseNodeKind::StrictEq:
|
|
case ParseNodeKind::Eq:
|
|
case ParseNodeKind::StrictNe:
|
|
case ParseNodeKind::Ne:
|
|
case ParseNodeKind::Lt:
|
|
case ParseNodeKind::Le:
|
|
case ParseNodeKind::Gt:
|
|
case ParseNodeKind::Ge:
|
|
case ParseNodeKind::InstanceOf:
|
|
case ParseNodeKind::In:
|
|
case ParseNodeKind::Lsh:
|
|
case ParseNodeKind::Rsh:
|
|
case ParseNodeKind::Ursh:
|
|
case ParseNodeKind::Add:
|
|
case ParseNodeKind::Sub:
|
|
case ParseNodeKind::Star:
|
|
case ParseNodeKind::Div:
|
|
case ParseNodeKind::Mod:
|
|
case ParseNodeKind::Pow:
|
|
case ParseNodeKind::Assign:
|
|
case ParseNodeKind::AddAssign:
|
|
case ParseNodeKind::SubAssign:
|
|
case ParseNodeKind::BitOrAssign:
|
|
case ParseNodeKind::BitXorAssign:
|
|
case ParseNodeKind::BitAndAssign:
|
|
case ParseNodeKind::LshAssign:
|
|
case ParseNodeKind::RshAssign:
|
|
case ParseNodeKind::UrshAssign:
|
|
case ParseNodeKind::MulAssign:
|
|
case ParseNodeKind::DivAssign:
|
|
case ParseNodeKind::ModAssign:
|
|
case ParseNodeKind::PowAssign:
|
|
case ParseNodeKind::Comma:
|
|
case ParseNodeKind::Array:
|
|
case ParseNodeKind::Object:
|
|
case ParseNodeKind::PropertyName:
|
|
case ParseNodeKind::Dot:
|
|
case ParseNodeKind::Elem:
|
|
case ParseNodeKind::Arguments:
|
|
case ParseNodeKind::Call:
|
|
case ParseNodeKind::Name:
|
|
case ParseNodeKind::TemplateString:
|
|
case ParseNodeKind::TemplateStringList:
|
|
case ParseNodeKind::TaggedTemplate:
|
|
case ParseNodeKind::CallSiteObj:
|
|
case ParseNodeKind::String:
|
|
case ParseNodeKind::RegExp:
|
|
case ParseNodeKind::True:
|
|
case ParseNodeKind::False:
|
|
case ParseNodeKind::Null:
|
|
case ParseNodeKind::RawUndefined:
|
|
case ParseNodeKind::This:
|
|
case ParseNodeKind::Elision:
|
|
case ParseNodeKind::Number:
|
|
case ParseNodeKind::New:
|
|
case ParseNodeKind::Generator:
|
|
case ParseNodeKind::ParamsBody:
|
|
case ParseNodeKind::Catch:
|
|
case ParseNodeKind::ForIn:
|
|
case ParseNodeKind::ForOf:
|
|
case ParseNodeKind::ForHead:
|
|
case ParseNodeKind::ClassMethod:
|
|
case ParseNodeKind::ClassMethodList:
|
|
case ParseNodeKind::ClassNames:
|
|
case ParseNodeKind::NewTarget:
|
|
case ParseNodeKind::ImportMeta:
|
|
case ParseNodeKind::PosHolder:
|
|
case ParseNodeKind::SuperCall:
|
|
case ParseNodeKind::SuperBase:
|
|
case ParseNodeKind::SetThis:
|
|
MOZ_CRASH("ContainsHoistedDeclaration should have indicated false on "
|
|
"some parent node without recurring to test this node");
|
|
|
|
case ParseNodeKind::Pipeline:
|
|
MOZ_ASSERT(node->isArity(PN_LIST));
|
|
*result = false;
|
|
return true;
|
|
|
|
case ParseNodeKind::Limit: // invalid sentinel value
|
|
MOZ_CRASH("unexpected ParseNodeKind::Limit in node");
|
|
}
|
|
|
|
MOZ_CRASH("invalid node kind");
|
|
}
|
|
|
|
/*
|
|
* Fold from one constant type to another.
|
|
* XXX handles only strings and numbers for now
|
|
*/
|
|
static bool
|
|
FoldType(JSContext* cx, ParseNode* pn, ParseNodeKind kind)
|
|
{
|
|
if (!pn->isKind(kind)) {
|
|
switch (kind) {
|
|
case ParseNodeKind::Number:
|
|
if (pn->isKind(ParseNodeKind::String)) {
|
|
double d;
|
|
if (!StringToNumber(cx, pn->pn_atom, &d))
|
|
return false;
|
|
pn->pn_dval = d;
|
|
pn->setKind(ParseNodeKind::Number);
|
|
pn->setOp(JSOP_DOUBLE);
|
|
}
|
|
break;
|
|
|
|
case ParseNodeKind::String:
|
|
if (pn->isKind(ParseNodeKind::Number)) {
|
|
pn->pn_atom = NumberToAtom(cx, pn->pn_dval);
|
|
if (!pn->pn_atom)
|
|
return false;
|
|
pn->setKind(ParseNodeKind::String);
|
|
pn->setOp(JSOP_STRING);
|
|
}
|
|
break;
|
|
|
|
default:;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
// Remove a ParseNode, **pnp, from a parse tree, putting another ParseNode,
|
|
// *pn, in its place.
|
|
//
|
|
// pnp points to a ParseNode pointer. This must be the only pointer that points
|
|
// to the parse node being replaced. The replacement, *pn, is unchanged except
|
|
// for its pn_next pointer; updating that is necessary if *pn's new parent is a
|
|
// list node.
|
|
static void
|
|
ReplaceNode(ParseNode** pnp, ParseNode* pn)
|
|
{
|
|
pn->pn_next = (*pnp)->pn_next;
|
|
*pnp = pn;
|
|
}
|
|
|
|
static bool
|
|
IsEffectless(ParseNode* node)
|
|
{
|
|
return node->isKind(ParseNodeKind::True) ||
|
|
node->isKind(ParseNodeKind::False) ||
|
|
node->isKind(ParseNodeKind::String) ||
|
|
node->isKind(ParseNodeKind::TemplateString) ||
|
|
node->isKind(ParseNodeKind::Number) ||
|
|
node->isKind(ParseNodeKind::Null) ||
|
|
node->isKind(ParseNodeKind::RawUndefined) ||
|
|
node->isKind(ParseNodeKind::Function);
|
|
}
|
|
|
|
enum Truthiness { Truthy, Falsy, Unknown };
|
|
|
|
static Truthiness
|
|
Boolish(ParseNode* pn)
|
|
{
|
|
switch (pn->getKind()) {
|
|
case ParseNodeKind::Number:
|
|
return (pn->pn_dval != 0 && !IsNaN(pn->pn_dval)) ? Truthy : Falsy;
|
|
|
|
case ParseNodeKind::String:
|
|
case ParseNodeKind::TemplateString:
|
|
return (pn->pn_atom->length() > 0) ? Truthy : Falsy;
|
|
|
|
case ParseNodeKind::True:
|
|
case ParseNodeKind::Function:
|
|
return Truthy;
|
|
|
|
case ParseNodeKind::False:
|
|
case ParseNodeKind::Null:
|
|
case ParseNodeKind::RawUndefined:
|
|
return Falsy;
|
|
|
|
case ParseNodeKind::Void: {
|
|
// |void <foo>| evaluates to |undefined| which isn't truthy. But the
|
|
// sense of this method requires that the expression be literally
|
|
// replaceable with true/false: not the case if the nested expression
|
|
// is effectful, might throw, &c. Walk past the |void| (and nested
|
|
// |void| expressions, for good measure) and check that the nested
|
|
// expression doesn't break this requirement before indicating falsity.
|
|
do {
|
|
pn = pn->pn_kid;
|
|
} while (pn->isKind(ParseNodeKind::Void));
|
|
|
|
return IsEffectless(pn) ? Falsy : Unknown;
|
|
}
|
|
|
|
default:
|
|
return Unknown;
|
|
}
|
|
}
|
|
|
|
static bool
|
|
Fold(JSContext* cx, ParseNode** pnp, PerHandlerParser<FullParseHandler>& parser);
|
|
|
|
static bool
|
|
FoldCondition(JSContext* cx, ParseNode** nodePtr, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
// Conditions fold like any other expression...
|
|
if (!Fold(cx, nodePtr, parser))
|
|
return false;
|
|
|
|
// ...but then they sometimes can be further folded to constants.
|
|
ParseNode* node = *nodePtr;
|
|
Truthiness t = Boolish(node);
|
|
if (t != Unknown) {
|
|
// We can turn function nodes into constant nodes here, but mutating
|
|
// function nodes is tricky --- in particular, mutating a function node
|
|
// that appears on a method list corrupts the method list. However,
|
|
// methods are M's in statements of the form 'this.foo = M;', which we
|
|
// never fold, so we're okay.
|
|
if (t == Truthy) {
|
|
node->setKind(ParseNodeKind::True);
|
|
node->setOp(JSOP_TRUE);
|
|
} else {
|
|
node->setKind(ParseNodeKind::False);
|
|
node->setOp(JSOP_FALSE);
|
|
}
|
|
node->setArity(PN_NULLARY);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldTypeOfExpr(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::TypeOfExpr));
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
|
|
ParseNode*& expr = node->pn_kid;
|
|
if (!Fold(cx, &expr, parser))
|
|
return false;
|
|
|
|
// Constant-fold the entire |typeof| if given a constant with known type.
|
|
RootedPropertyName result(cx);
|
|
if (expr->isKind(ParseNodeKind::String) || expr->isKind(ParseNodeKind::TemplateString))
|
|
result = cx->names().string;
|
|
else if (expr->isKind(ParseNodeKind::Number))
|
|
result = cx->names().number;
|
|
else if (expr->isKind(ParseNodeKind::Null))
|
|
result = cx->names().object;
|
|
else if (expr->isKind(ParseNodeKind::True) || expr->isKind(ParseNodeKind::False))
|
|
result = cx->names().boolean;
|
|
else if (expr->isKind(ParseNodeKind::Function))
|
|
result = cx->names().function;
|
|
|
|
if (result) {
|
|
node->setKind(ParseNodeKind::String);
|
|
node->setArity(PN_NULLARY);
|
|
node->setOp(JSOP_NOP);
|
|
node->pn_atom = result;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldDeleteExpr(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::DeleteExpr));
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
|
|
ParseNode*& expr = node->pn_kid;
|
|
if (!Fold(cx, &expr, parser))
|
|
return false;
|
|
|
|
// Expression deletion evaluates the expression, then evaluates to true.
|
|
// For effectless expressions, eliminate the expression evaluation.
|
|
if (IsEffectless(expr)) {
|
|
node->setKind(ParseNodeKind::True);
|
|
node->setArity(PN_NULLARY);
|
|
node->setOp(JSOP_TRUE);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldDeleteElement(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::DeleteElem));
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
MOZ_ASSERT(node->pn_kid->isKind(ParseNodeKind::Elem));
|
|
|
|
ParseNode*& expr = node->pn_kid;
|
|
if (!Fold(cx, &expr, parser))
|
|
return false;
|
|
|
|
// If we're deleting an element, but constant-folding converted our
|
|
// element reference into a dotted property access, we must *also*
|
|
// morph the node's kind.
|
|
//
|
|
// In principle this also applies to |super["foo"] -> super.foo|,
|
|
// but we don't constant-fold |super["foo"]| yet.
|
|
MOZ_ASSERT(expr->isKind(ParseNodeKind::Elem) || expr->isKind(ParseNodeKind::Dot));
|
|
if (expr->isKind(ParseNodeKind::Dot))
|
|
node->setKind(ParseNodeKind::DeleteProp);
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldDeleteProperty(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::DeleteProp));
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
MOZ_ASSERT(node->pn_kid->isKind(ParseNodeKind::Dot));
|
|
|
|
ParseNode*& expr = node->pn_kid;
|
|
#ifdef DEBUG
|
|
ParseNodeKind oldKind = expr->getKind();
|
|
#endif
|
|
|
|
if (!Fold(cx, &expr, parser))
|
|
return false;
|
|
|
|
MOZ_ASSERT(expr->isKind(oldKind),
|
|
"kind should have remained invariant under folding");
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldNot(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Not));
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
|
|
ParseNode*& expr = node->pn_kid;
|
|
if (!FoldCondition(cx, &expr, parser))
|
|
return false;
|
|
|
|
if (expr->isKind(ParseNodeKind::Number)) {
|
|
double d = expr->pn_dval;
|
|
|
|
if (d == 0 || IsNaN(d)) {
|
|
node->setKind(ParseNodeKind::True);
|
|
node->setOp(JSOP_TRUE);
|
|
} else {
|
|
node->setKind(ParseNodeKind::False);
|
|
node->setOp(JSOP_FALSE);
|
|
}
|
|
node->setArity(PN_NULLARY);
|
|
} else if (expr->isKind(ParseNodeKind::True) || expr->isKind(ParseNodeKind::False)) {
|
|
bool newval = !expr->isKind(ParseNodeKind::True);
|
|
|
|
node->setKind(newval ? ParseNodeKind::True : ParseNodeKind::False);
|
|
node->setArity(PN_NULLARY);
|
|
node->setOp(newval ? JSOP_TRUE : JSOP_FALSE);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldUnaryArithmetic(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::BitNot) ||
|
|
node->isKind(ParseNodeKind::Pos) ||
|
|
node->isKind(ParseNodeKind::Neg),
|
|
"need a different method for this node kind");
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
|
|
ParseNode*& expr = node->pn_kid;
|
|
if (!Fold(cx, &expr, parser))
|
|
return false;
|
|
|
|
if (expr->isKind(ParseNodeKind::Number) ||
|
|
expr->isKind(ParseNodeKind::True) ||
|
|
expr->isKind(ParseNodeKind::False))
|
|
{
|
|
double d = expr->isKind(ParseNodeKind::Number)
|
|
? expr->pn_dval
|
|
: double(expr->isKind(ParseNodeKind::True));
|
|
|
|
if (node->isKind(ParseNodeKind::BitNot))
|
|
d = ~ToInt32(d);
|
|
else if (node->isKind(ParseNodeKind::Neg))
|
|
d = -d;
|
|
else
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Pos)); // nothing to do
|
|
|
|
node->setKind(ParseNodeKind::Number);
|
|
node->setOp(JSOP_DOUBLE);
|
|
node->setArity(PN_NULLARY);
|
|
node->pn_dval = d;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldIncrementDecrement(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::PreIncrement) ||
|
|
node->isKind(ParseNodeKind::PostIncrement) ||
|
|
node->isKind(ParseNodeKind::PreDecrement) ||
|
|
node->isKind(ParseNodeKind::PostDecrement));
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
|
|
ParseNode*& target = node->pn_kid;
|
|
MOZ_ASSERT(parser.isValidSimpleAssignmentTarget(target, PermitAssignmentToFunctionCalls));
|
|
|
|
if (!Fold(cx, &target, parser))
|
|
return false;
|
|
|
|
MOZ_ASSERT(parser.isValidSimpleAssignmentTarget(target, PermitAssignmentToFunctionCalls));
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldAndOr(JSContext* cx, ParseNode** nodePtr, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
ParseNode* node = *nodePtr;
|
|
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::And) || node->isKind(ParseNodeKind::Or));
|
|
MOZ_ASSERT(node->isArity(PN_LIST));
|
|
|
|
bool isOrNode = node->isKind(ParseNodeKind::Or);
|
|
ParseNode** elem = &node->pn_head;
|
|
do {
|
|
if (!Fold(cx, elem, parser))
|
|
return false;
|
|
|
|
Truthiness t = Boolish(*elem);
|
|
|
|
// If we don't know the constant-folded node's truthiness, we can't
|
|
// reduce this node with its surroundings. Continue folding any
|
|
// remaining nodes.
|
|
if (t == Unknown) {
|
|
elem = &(*elem)->pn_next;
|
|
continue;
|
|
}
|
|
|
|
// If the constant-folded node's truthiness will terminate the
|
|
// condition -- `a || true || expr` or |b && false && expr| -- then
|
|
// trailing nodes will never be evaluated. Truncate the list after
|
|
// the known-truthiness node, as it's the overall result.
|
|
if ((t == Truthy) == isOrNode) {
|
|
for (ParseNode* next = (*elem)->pn_next; next; next = next->pn_next)
|
|
--node->pn_count;
|
|
|
|
// Terminate the original and/or list at the known-truthiness
|
|
// node.
|
|
(*elem)->pn_next = nullptr;
|
|
elem = &(*elem)->pn_next;
|
|
break;
|
|
}
|
|
|
|
MOZ_ASSERT((t == Truthy) == !isOrNode);
|
|
|
|
// We've encountered a vacuous node that'll never short-circuit
|
|
// evaluation.
|
|
if ((*elem)->pn_next) {
|
|
// This node is never the overall result when there are
|
|
// subsequent nodes. Remove it.
|
|
ParseNode* elt = *elem;
|
|
*elem = elt->pn_next;
|
|
--node->pn_count;
|
|
} else {
|
|
// Otherwise this node is the result of the overall expression,
|
|
// so leave it alone. And we're done.
|
|
elem = &(*elem)->pn_next;
|
|
break;
|
|
}
|
|
} while (*elem);
|
|
|
|
// If the last node in the list was replaced, we need to update the
|
|
// tail pointer in the original and/or node.
|
|
node->pn_tail = elem;
|
|
|
|
node->checkListConsistency();
|
|
|
|
// If we removed nodes, we may have to replace a one-element list with
|
|
// its element.
|
|
if (node->pn_count == 1) {
|
|
ParseNode* first = node->pn_head;
|
|
ReplaceNode(nodePtr, first);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldConditional(JSContext* cx, ParseNode** nodePtr, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
ParseNode** nextNode = nodePtr;
|
|
|
|
do {
|
|
// |nextNode| on entry points to the C?T:F expression to be folded.
|
|
// Reset it to exit the loop in the common case where F isn't another
|
|
// ?: expression.
|
|
nodePtr = nextNode;
|
|
nextNode = nullptr;
|
|
|
|
ParseNode* node = *nodePtr;
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Conditional));
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
|
|
ParseNode*& expr = node->pn_kid1;
|
|
if (!FoldCondition(cx, &expr, parser))
|
|
return false;
|
|
|
|
ParseNode*& ifTruthy = node->pn_kid2;
|
|
if (!Fold(cx, &ifTruthy, parser))
|
|
return false;
|
|
|
|
ParseNode*& ifFalsy = node->pn_kid3;
|
|
|
|
// If our C?T:F node has F as another ?: node, *iteratively* constant-
|
|
// fold F *after* folding C and T (and possibly eliminating C and one
|
|
// of T/F entirely); otherwise fold F normally. Making |nextNode| non-
|
|
// null causes this loop to run again to fold F.
|
|
//
|
|
// Conceivably we could instead/also iteratively constant-fold T, if T
|
|
// were more complex than F. Such an optimization is unimplemented.
|
|
if (ifFalsy->isKind(ParseNodeKind::Conditional)) {
|
|
nextNode = &ifFalsy;
|
|
} else {
|
|
if (!Fold(cx, &ifFalsy, parser))
|
|
return false;
|
|
}
|
|
|
|
// Try to constant-fold based on the condition expression.
|
|
Truthiness t = Boolish(expr);
|
|
if (t == Unknown)
|
|
continue;
|
|
|
|
// Otherwise reduce 'C ? T : F' to T or F as directed by C.
|
|
ParseNode* replacement = t == Truthy ? ifTruthy : ifFalsy;
|
|
|
|
// Otherwise perform a replacement. This invalidates |nextNode|, so
|
|
// reset it (if the replacement requires folding) or clear it (if
|
|
// |ifFalsy| is dead code) as needed.
|
|
if (nextNode)
|
|
nextNode = (*nextNode == replacement) ? nodePtr : nullptr;
|
|
ReplaceNode(nodePtr, replacement);
|
|
} while (nextNode);
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldIf(JSContext* cx, ParseNode** nodePtr, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
ParseNode** nextNode = nodePtr;
|
|
|
|
do {
|
|
// |nextNode| on entry points to the initial |if| to be folded. Reset
|
|
// it to exit the loop when the |else| arm isn't another |if|.
|
|
nodePtr = nextNode;
|
|
nextNode = nullptr;
|
|
|
|
ParseNode* node = *nodePtr;
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::If));
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
|
|
ParseNode*& expr = node->pn_kid1;
|
|
if (!FoldCondition(cx, &expr, parser))
|
|
return false;
|
|
|
|
ParseNode*& consequent = node->pn_kid2;
|
|
if (!Fold(cx, &consequent, parser))
|
|
return false;
|
|
|
|
ParseNode*& alternative = node->pn_kid3;
|
|
if (alternative) {
|
|
// If in |if (C) T; else F;| we have |F| as another |if|,
|
|
// *iteratively* constant-fold |F| *after* folding |C| and |T| (and
|
|
// possibly completely replacing the whole thing with |T| or |F|);
|
|
// otherwise fold F normally. Making |nextNode| non-null causes
|
|
// this loop to run again to fold F.
|
|
if (alternative->isKind(ParseNodeKind::If)) {
|
|
nextNode = &alternative;
|
|
} else {
|
|
if (!Fold(cx, &alternative, parser))
|
|
return false;
|
|
}
|
|
}
|
|
|
|
// Eliminate the consequent or alternative if the condition has
|
|
// constant truthiness.
|
|
Truthiness t = Boolish(expr);
|
|
if (t == Unknown)
|
|
continue;
|
|
|
|
// Careful! Either of these can be null: |replacement| in |if (0) T;|,
|
|
// and |discarded| in |if (true) T;|.
|
|
ParseNode* replacement;
|
|
ParseNode* discarded;
|
|
if (t == Truthy) {
|
|
replacement = consequent;
|
|
discarded = alternative;
|
|
} else {
|
|
replacement = alternative;
|
|
discarded = consequent;
|
|
}
|
|
|
|
bool performReplacement = true;
|
|
if (discarded) {
|
|
// A declaration that hoists outside the discarded arm prevents the
|
|
// |if| from being folded away.
|
|
bool containsHoistedDecls;
|
|
if (!ContainsHoistedDeclaration(cx, discarded, &containsHoistedDecls))
|
|
return false;
|
|
|
|
performReplacement = !containsHoistedDecls;
|
|
}
|
|
|
|
if (!performReplacement)
|
|
continue;
|
|
|
|
if (!replacement) {
|
|
// If there's no replacement node, we have a constantly-false |if|
|
|
// with no |else|. Replace the entire thing with an empty
|
|
// statement list.
|
|
node->setKind(ParseNodeKind::StatementList);
|
|
node->setArity(PN_LIST);
|
|
node->makeEmpty();
|
|
} else {
|
|
// Replacement invalidates |nextNode|, so reset it (if the
|
|
// replacement requires folding) or clear it (if |alternative|
|
|
// is dead code) as needed.
|
|
if (nextNode)
|
|
nextNode = (*nextNode == replacement) ? nodePtr : nullptr;
|
|
ReplaceNode(nodePtr, replacement);
|
|
}
|
|
} while (nextNode);
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldFunction(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Function));
|
|
MOZ_ASSERT(node->isArity(PN_CODE));
|
|
|
|
// Don't constant-fold inside "use asm" code, as this could create a parse
|
|
// tree that doesn't type-check as asm.js.
|
|
if (node->pn_funbox->useAsmOrInsideUseAsm())
|
|
return true;
|
|
|
|
// Note: pn_body is null for lazily-parsed functions.
|
|
if (ParseNode*& functionBody = node->pn_body) {
|
|
if (!Fold(cx, &functionBody, parser))
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static double
|
|
ComputeBinary(ParseNodeKind kind, double left, double right)
|
|
{
|
|
if (kind == ParseNodeKind::Add)
|
|
return left + right;
|
|
|
|
if (kind == ParseNodeKind::Sub)
|
|
return left - right;
|
|
|
|
if (kind == ParseNodeKind::Star)
|
|
return left * right;
|
|
|
|
if (kind == ParseNodeKind::Mod)
|
|
return NumberMod(left, right);
|
|
|
|
if (kind == ParseNodeKind::Ursh)
|
|
return ToUint32(left) >> (ToUint32(right) & 31);
|
|
|
|
if (kind == ParseNodeKind::Div)
|
|
return NumberDiv(left, right);
|
|
|
|
MOZ_ASSERT(kind == ParseNodeKind::Lsh || kind == ParseNodeKind::Rsh);
|
|
|
|
int32_t i = ToInt32(left);
|
|
uint32_t j = ToUint32(right) & 31;
|
|
return int32_t((kind == ParseNodeKind::Lsh) ? uint32_t(i) << j : i >> j);
|
|
}
|
|
|
|
static bool
|
|
FoldModule(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Module));
|
|
MOZ_ASSERT(node->isArity(PN_CODE));
|
|
|
|
ParseNode*& moduleBody = node->pn_body;
|
|
MOZ_ASSERT(moduleBody);
|
|
return Fold(cx, &moduleBody, parser);
|
|
}
|
|
|
|
static bool
|
|
FoldBinaryArithmetic(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Sub) ||
|
|
node->isKind(ParseNodeKind::Star) ||
|
|
node->isKind(ParseNodeKind::Lsh) ||
|
|
node->isKind(ParseNodeKind::Rsh) ||
|
|
node->isKind(ParseNodeKind::Ursh) ||
|
|
node->isKind(ParseNodeKind::Div) ||
|
|
node->isKind(ParseNodeKind::Mod));
|
|
MOZ_ASSERT(node->isArity(PN_LIST));
|
|
MOZ_ASSERT(node->pn_count >= 2);
|
|
|
|
// Fold each operand, ideally into a number.
|
|
ParseNode** listp = &node->pn_head;
|
|
for (; *listp; listp = &(*listp)->pn_next) {
|
|
if (!Fold(cx, listp, parser))
|
|
return false;
|
|
|
|
if (!FoldType(cx, *listp, ParseNodeKind::Number))
|
|
return false;
|
|
}
|
|
|
|
// Repoint the list's tail pointer.
|
|
node->pn_tail = listp;
|
|
|
|
// Now fold all leading numeric terms together into a single number.
|
|
// (Trailing terms for the non-shift operations can't be folded together
|
|
// due to floating point imprecision. For example, if |x === -2**53|,
|
|
// |x - 1 - 1 === -2**53| but |x - 2 === -2**53 - 2|. Shifts could be
|
|
// folded, but it doesn't seem worth the effort.)
|
|
ParseNode* elem = node->pn_head;
|
|
ParseNode* next = elem->pn_next;
|
|
if (elem->isKind(ParseNodeKind::Number)) {
|
|
ParseNodeKind kind = node->getKind();
|
|
while (true) {
|
|
if (!next || !next->isKind(ParseNodeKind::Number))
|
|
break;
|
|
|
|
double d = ComputeBinary(kind, elem->pn_dval, next->pn_dval);
|
|
|
|
next = next->pn_next;
|
|
elem->pn_next = next;
|
|
|
|
elem->setKind(ParseNodeKind::Number);
|
|
elem->setOp(JSOP_DOUBLE);
|
|
elem->setArity(PN_NULLARY);
|
|
elem->pn_dval = d;
|
|
|
|
node->pn_count--;
|
|
}
|
|
|
|
if (node->pn_count == 1) {
|
|
MOZ_ASSERT(node->pn_head == elem);
|
|
MOZ_ASSERT(elem->isKind(ParseNodeKind::Number));
|
|
|
|
double d = elem->pn_dval;
|
|
node->setKind(ParseNodeKind::Number);
|
|
node->setArity(PN_NULLARY);
|
|
node->setOp(JSOP_DOUBLE);
|
|
node->pn_dval = d;
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldExponentiation(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Pow));
|
|
MOZ_ASSERT(node->isArity(PN_LIST));
|
|
MOZ_ASSERT(node->pn_count >= 2);
|
|
|
|
// Fold each operand, ideally into a number.
|
|
ParseNode** listp = &node->pn_head;
|
|
for (; *listp; listp = &(*listp)->pn_next) {
|
|
if (!Fold(cx, listp, parser))
|
|
return false;
|
|
|
|
if (!FoldType(cx, *listp, ParseNodeKind::Number))
|
|
return false;
|
|
}
|
|
|
|
// Repoint the list's tail pointer.
|
|
node->pn_tail = listp;
|
|
|
|
// Unlike all other binary arithmetic operators, ** is right-associative:
|
|
// 2**3**5 is 2**(3**5), not (2**3)**5. As list nodes singly-link their
|
|
// children, full constant-folding requires either linear space or dodgy
|
|
// in-place linked list reversal. So we only fold one exponentiation: it's
|
|
// easy and addresses common cases like |2**32|.
|
|
if (node->pn_count > 2)
|
|
return true;
|
|
|
|
ParseNode* base = node->pn_head;
|
|
ParseNode* exponent = base->pn_next;
|
|
if (!base->isKind(ParseNodeKind::Number) || !exponent->isKind(ParseNodeKind::Number))
|
|
return true;
|
|
|
|
double d1 = base->pn_dval, d2 = exponent->pn_dval;
|
|
|
|
node->setKind(ParseNodeKind::Number);
|
|
node->setArity(PN_NULLARY);
|
|
node->setOp(JSOP_DOUBLE);
|
|
node->pn_dval = ecmaPow(d1, d2);
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldList(JSContext* cx, ParseNode* list, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(list->isArity(PN_LIST));
|
|
|
|
ParseNode** elem = &list->pn_head;
|
|
for (; *elem; elem = &(*elem)->pn_next) {
|
|
if (!Fold(cx, elem, parser))
|
|
return false;
|
|
}
|
|
|
|
// Repoint the list's tail pointer if the final element was replaced.
|
|
list->pn_tail = elem;
|
|
|
|
list->checkListConsistency();
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldReturn(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Return));
|
|
MOZ_ASSERT(node->isArity(PN_UNARY));
|
|
|
|
if (ParseNode*& expr = node->pn_kid) {
|
|
if (!Fold(cx, &expr, parser))
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldTry(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Try));
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
|
|
ParseNode*& statements = node->pn_kid1;
|
|
if (!Fold(cx, &statements, parser))
|
|
return false;
|
|
|
|
if (ParseNode*& catchScope = node->pn_kid2) {
|
|
if (!Fold(cx, &catchScope, parser))
|
|
return false;
|
|
}
|
|
|
|
if (ParseNode*& finally = node->pn_kid3) {
|
|
if (!Fold(cx, &finally, parser))
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldCatch(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Catch));
|
|
MOZ_ASSERT(node->isArity(PN_BINARY));
|
|
|
|
if (ParseNode*& declPattern = node->pn_left) {
|
|
if (!Fold(cx, &declPattern, parser))
|
|
return false;
|
|
}
|
|
|
|
if (ParseNode*& statements = node->pn_right) {
|
|
if (!Fold(cx, &statements, parser))
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldClass(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Class));
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
|
|
if (ParseNode*& classNames = node->pn_kid1) {
|
|
if (!Fold(cx, &classNames, parser))
|
|
return false;
|
|
}
|
|
|
|
if (ParseNode*& heritage = node->pn_kid2) {
|
|
if (!Fold(cx, &heritage, parser))
|
|
return false;
|
|
}
|
|
|
|
ParseNode*& body = node->pn_kid3;
|
|
return Fold(cx, &body, parser);
|
|
}
|
|
|
|
static bool
|
|
FoldElement(JSContext* cx, ParseNode** nodePtr, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
ParseNode* node = *nodePtr;
|
|
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Elem));
|
|
MOZ_ASSERT(node->isArity(PN_BINARY));
|
|
|
|
ParseNode*& expr = node->pn_left;
|
|
if (!Fold(cx, &expr, parser))
|
|
return false;
|
|
|
|
ParseNode*& key = node->pn_right;
|
|
if (!Fold(cx, &key, parser))
|
|
return false;
|
|
|
|
PropertyName* name = nullptr;
|
|
if (key->isKind(ParseNodeKind::String)) {
|
|
JSAtom* atom = key->pn_atom;
|
|
uint32_t index;
|
|
|
|
if (atom->isIndex(&index)) {
|
|
// Optimization 1: We have something like expr["100"]. This is
|
|
// equivalent to expr[100] which is faster.
|
|
key->setKind(ParseNodeKind::Number);
|
|
key->setOp(JSOP_DOUBLE);
|
|
key->pn_dval = index;
|
|
} else {
|
|
name = atom->asPropertyName();
|
|
}
|
|
} else if (key->isKind(ParseNodeKind::Number)) {
|
|
double number = key->pn_dval;
|
|
if (number != ToUint32(number)) {
|
|
// Optimization 2: We have something like expr[3.14]. The number
|
|
// isn't an array index, so it converts to a string ("3.14"),
|
|
// enabling optimization 3 below.
|
|
JSAtom* atom = NumberToAtom(cx, number);
|
|
if (!atom)
|
|
return false;
|
|
name = atom->asPropertyName();
|
|
}
|
|
}
|
|
|
|
// If we don't have a name, we can't optimize to getprop.
|
|
if (!name)
|
|
return true;
|
|
|
|
// Optimization 3: We have expr["foo"] where foo is not an index. Convert
|
|
// to a property access (like expr.foo) that optimizes better downstream.
|
|
ParseNode* nameNode = parser.newPropertyName(name, key->pn_pos);
|
|
if (!nameNode)
|
|
return false;
|
|
ParseNode* dottedAccess = parser.newPropertyAccess(expr, nameNode);
|
|
if (!dottedAccess)
|
|
return false;
|
|
dottedAccess->setInParens(node->isInParens());
|
|
ReplaceNode(nodePtr, dottedAccess);
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldAdd(JSContext* cx, ParseNode** nodePtr, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
ParseNode* node = *nodePtr;
|
|
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Add));
|
|
MOZ_ASSERT(node->isArity(PN_LIST));
|
|
MOZ_ASSERT(node->pn_count >= 2);
|
|
|
|
// Generically fold all operands first.
|
|
if (!FoldList(cx, node, parser))
|
|
return false;
|
|
|
|
// Fold leading numeric operands together:
|
|
//
|
|
// (1 + 2 + x) becomes (3 + x)
|
|
//
|
|
// Don't go past the leading operands: additions after a string are
|
|
// string concatenations, not additions: ("1" + 2 + 3 === "123").
|
|
ParseNode* current = node->pn_head;
|
|
ParseNode* next = current->pn_next;
|
|
if (current->isKind(ParseNodeKind::Number)) {
|
|
do {
|
|
if (!next->isKind(ParseNodeKind::Number))
|
|
break;
|
|
|
|
current->pn_dval += next->pn_dval;
|
|
current->pn_next = next->pn_next;
|
|
next = current->pn_next;
|
|
|
|
MOZ_ASSERT(node->pn_count > 1);
|
|
node->pn_count--;
|
|
} while (next);
|
|
}
|
|
|
|
// If any operands remain, attempt string concatenation folding.
|
|
do {
|
|
// If no operands remain, we're done.
|
|
if (!next)
|
|
break;
|
|
|
|
// (number + string) is string concatenation *only* at the start of
|
|
// the list: (x + 1 + "2" !== x + "12") when x is a number.
|
|
if (current->isKind(ParseNodeKind::Number) && next->isKind(ParseNodeKind::String)) {
|
|
if (!FoldType(cx, current, ParseNodeKind::String))
|
|
return false;
|
|
next = current->pn_next;
|
|
}
|
|
|
|
// The first string forces all subsequent additions to be
|
|
// string concatenations.
|
|
do {
|
|
if (current->isKind(ParseNodeKind::String))
|
|
break;
|
|
|
|
current = next;
|
|
next = next->pn_next;
|
|
} while (next);
|
|
|
|
// If there's nothing left to fold, we're done.
|
|
if (!next)
|
|
break;
|
|
|
|
RootedString combination(cx);
|
|
RootedString tmp(cx);
|
|
do {
|
|
// Create a rope of the current string and all succeeding
|
|
// constants that we can convert to strings, then atomize it
|
|
// and replace them all with that fresh string.
|
|
MOZ_ASSERT(current->isKind(ParseNodeKind::String));
|
|
|
|
combination = current->pn_atom;
|
|
|
|
do {
|
|
// Try folding the next operand to a string.
|
|
if (!FoldType(cx, next, ParseNodeKind::String))
|
|
return false;
|
|
|
|
// Stop glomming once folding doesn't produce a string.
|
|
if (!next->isKind(ParseNodeKind::String))
|
|
break;
|
|
|
|
// Add this string to the combination and remove the node.
|
|
tmp = next->pn_atom;
|
|
combination = ConcatStrings<CanGC>(cx, combination, tmp);
|
|
if (!combination)
|
|
return false;
|
|
|
|
next = next->pn_next;
|
|
current->pn_next = next;
|
|
|
|
MOZ_ASSERT(node->pn_count > 1);
|
|
node->pn_count--;
|
|
} while (next);
|
|
|
|
// Replace |current|'s string with the entire combination.
|
|
MOZ_ASSERT(current->isKind(ParseNodeKind::String));
|
|
combination = AtomizeString(cx, combination);
|
|
if (!combination)
|
|
return false;
|
|
current->pn_atom = &combination->asAtom();
|
|
|
|
|
|
// If we're out of nodes, we're done.
|
|
if (!next)
|
|
break;
|
|
|
|
current = next;
|
|
next = current->pn_next;
|
|
|
|
// If we're out of nodes *after* the non-foldable-to-string
|
|
// node, we're done.
|
|
if (!next)
|
|
break;
|
|
|
|
// Otherwise find the next node foldable to a string, and loop.
|
|
do {
|
|
current = next;
|
|
next = current->pn_next;
|
|
|
|
if (!FoldType(cx, current, ParseNodeKind::String))
|
|
return false;
|
|
next = current->pn_next;
|
|
} while (!current->isKind(ParseNodeKind::String) && next);
|
|
} while (next);
|
|
} while (false);
|
|
|
|
MOZ_ASSERT(!next, "must have considered all nodes here");
|
|
MOZ_ASSERT(!current->pn_next, "current node must be the last node");
|
|
|
|
node->pn_tail = ¤t->pn_next;
|
|
node->checkListConsistency();
|
|
|
|
if (node->pn_count == 1) {
|
|
// We reduced the list to a constant. Replace the ParseNodeKind::Add node
|
|
// with that constant.
|
|
ReplaceNode(nodePtr, current);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldCall(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Call) ||
|
|
node->isKind(ParseNodeKind::SuperCall) ||
|
|
node->isKind(ParseNodeKind::New) ||
|
|
node->isKind(ParseNodeKind::TaggedTemplate));
|
|
MOZ_ASSERT(node->isArity(PN_BINARY));
|
|
|
|
// Don't fold a parenthesized callable component in an invocation, as this
|
|
// might cause a different |this| value to be used, changing semantics:
|
|
//
|
|
// var prop = "global";
|
|
// var obj = { prop: "obj", f: function() { return this.prop; } };
|
|
// assertEq((true ? obj.f : null)(), "global");
|
|
// assertEq(obj.f(), "obj");
|
|
// assertEq((true ? obj.f : null)``, "global");
|
|
// assertEq(obj.f``, "obj");
|
|
//
|
|
// See bug 537673 and bug 1182373.
|
|
ParseNode** pn_callee = &node->pn_left;
|
|
if (node->isKind(ParseNodeKind::New) || !(*pn_callee)->isInParens()) {
|
|
if (!Fold(cx, pn_callee, parser))
|
|
return false;
|
|
}
|
|
|
|
ParseNode** pn_args = &node->pn_right;
|
|
if (!Fold(cx, pn_args, parser))
|
|
return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldArguments(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Arguments));
|
|
MOZ_ASSERT(node->isArity(PN_LIST));
|
|
|
|
ParseNode** listp = &node->pn_head;
|
|
for (; *listp; listp = &(*listp)->pn_next) {
|
|
if (!Fold(cx, listp, parser))
|
|
return false;
|
|
}
|
|
|
|
// If the last node in the list was replaced, pn_tail points into the wrong node.
|
|
node->pn_tail = listp;
|
|
|
|
node->checkListConsistency();
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldForInOrOf(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::ForIn) ||
|
|
node->isKind(ParseNodeKind::ForOf));
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
MOZ_ASSERT(!node->pn_kid2);
|
|
|
|
return Fold(cx, &node->pn_kid1, parser) &&
|
|
Fold(cx, &node->pn_kid3, parser);
|
|
}
|
|
|
|
static bool
|
|
FoldForHead(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::ForHead));
|
|
MOZ_ASSERT(node->isArity(PN_TERNARY));
|
|
|
|
if (ParseNode*& init = node->pn_kid1) {
|
|
if (!Fold(cx, &init, parser))
|
|
return false;
|
|
}
|
|
|
|
if (ParseNode*& test = node->pn_kid2) {
|
|
if (!FoldCondition(cx, &test, parser))
|
|
return false;
|
|
|
|
if (test->isKind(ParseNodeKind::True))
|
|
test = nullptr;
|
|
}
|
|
|
|
if (ParseNode*& update = node->pn_kid3) {
|
|
if (!Fold(cx, &update, parser))
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
FoldDottedProperty(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Dot));
|
|
MOZ_ASSERT(node->isArity(PN_BINARY));
|
|
|
|
// Iterate through a long chain of dotted property accesses to find the
|
|
// most-nested non-dotted property node, then fold that.
|
|
ParseNode** nested = &node->pn_left;
|
|
while ((*nested)->isKind(ParseNodeKind::Dot)) {
|
|
MOZ_ASSERT((*nested)->isArity(PN_BINARY));
|
|
nested = &(*nested)->pn_left;
|
|
}
|
|
|
|
return Fold(cx, nested, parser);
|
|
}
|
|
|
|
static bool
|
|
FoldName(JSContext* cx, ParseNode* node, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
MOZ_ASSERT(node->isKind(ParseNodeKind::Name));
|
|
MOZ_ASSERT(node->isArity(PN_NAME));
|
|
|
|
if (!node->pn_expr)
|
|
return true;
|
|
|
|
return Fold(cx, &node->pn_expr, parser);
|
|
}
|
|
|
|
bool
|
|
Fold(JSContext* cx, ParseNode** pnp, PerHandlerParser<FullParseHandler>& parser)
|
|
{
|
|
if (!CheckRecursionLimit(cx))
|
|
return false;
|
|
|
|
ParseNode* pn = *pnp;
|
|
|
|
switch (pn->getKind()) {
|
|
case ParseNodeKind::EmptyStatement:
|
|
case ParseNodeKind::RegExp:
|
|
case ParseNodeKind::String:
|
|
case ParseNodeKind::True:
|
|
case ParseNodeKind::False:
|
|
case ParseNodeKind::Null:
|
|
case ParseNodeKind::RawUndefined:
|
|
case ParseNodeKind::Elision:
|
|
case ParseNodeKind::Number:
|
|
case ParseNodeKind::Debugger:
|
|
case ParseNodeKind::Break:
|
|
case ParseNodeKind::Continue:
|
|
case ParseNodeKind::TemplateString:
|
|
case ParseNodeKind::Generator:
|
|
case ParseNodeKind::ExportBatchSpec:
|
|
case ParseNodeKind::ObjectPropertyName:
|
|
case ParseNodeKind::PosHolder:
|
|
MOZ_ASSERT(pn->isArity(PN_NULLARY));
|
|
return true;
|
|
|
|
case ParseNodeKind::SuperBase:
|
|
case ParseNodeKind::TypeOfName:
|
|
MOZ_ASSERT(pn->isArity(PN_UNARY));
|
|
MOZ_ASSERT(pn->pn_kid->isKind(ParseNodeKind::Name));
|
|
MOZ_ASSERT(!pn->pn_kid->expr());
|
|
return true;
|
|
|
|
case ParseNodeKind::TypeOfExpr:
|
|
return FoldTypeOfExpr(cx, pn, parser);
|
|
|
|
case ParseNodeKind::DeleteName: {
|
|
MOZ_ASSERT(pn->isArity(PN_UNARY));
|
|
MOZ_ASSERT(pn->pn_kid->isKind(ParseNodeKind::Name));
|
|
return true;
|
|
}
|
|
|
|
case ParseNodeKind::DeleteExpr:
|
|
return FoldDeleteExpr(cx, pn, parser);
|
|
|
|
case ParseNodeKind::DeleteElem:
|
|
return FoldDeleteElement(cx, pn, parser);
|
|
|
|
case ParseNodeKind::DeleteProp:
|
|
return FoldDeleteProperty(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Conditional:
|
|
return FoldConditional(cx, pnp, parser);
|
|
|
|
case ParseNodeKind::If:
|
|
return FoldIf(cx, pnp, parser);
|
|
|
|
case ParseNodeKind::Not:
|
|
return FoldNot(cx, pn, parser);
|
|
|
|
case ParseNodeKind::BitNot:
|
|
case ParseNodeKind::Pos:
|
|
case ParseNodeKind::Neg:
|
|
return FoldUnaryArithmetic(cx, pn, parser);
|
|
|
|
case ParseNodeKind::PreIncrement:
|
|
case ParseNodeKind::PostIncrement:
|
|
case ParseNodeKind::PreDecrement:
|
|
case ParseNodeKind::PostDecrement:
|
|
return FoldIncrementDecrement(cx, pn, parser);
|
|
|
|
case ParseNodeKind::ExpressionStatement:
|
|
case ParseNodeKind::Throw:
|
|
case ParseNodeKind::MutateProto:
|
|
case ParseNodeKind::ComputedName:
|
|
case ParseNodeKind::Spread:
|
|
case ParseNodeKind::Export:
|
|
case ParseNodeKind::Void:
|
|
MOZ_ASSERT(pn->isArity(PN_UNARY));
|
|
return Fold(cx, &pn->pn_kid, parser);
|
|
|
|
case ParseNodeKind::ExportDefault:
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
return Fold(cx, &pn->pn_left, parser);
|
|
|
|
case ParseNodeKind::This:
|
|
MOZ_ASSERT(pn->isArity(PN_UNARY));
|
|
if (ParseNode*& expr = pn->pn_kid)
|
|
return Fold(cx, &expr, parser);
|
|
return true;
|
|
|
|
case ParseNodeKind::Pipeline:
|
|
return true;
|
|
|
|
case ParseNodeKind::And:
|
|
case ParseNodeKind::Or:
|
|
return FoldAndOr(cx, pnp, parser);
|
|
|
|
case ParseNodeKind::Function:
|
|
return FoldFunction(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Module:
|
|
return FoldModule(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Sub:
|
|
case ParseNodeKind::Star:
|
|
case ParseNodeKind::Lsh:
|
|
case ParseNodeKind::Rsh:
|
|
case ParseNodeKind::Ursh:
|
|
case ParseNodeKind::Div:
|
|
case ParseNodeKind::Mod:
|
|
return FoldBinaryArithmetic(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Pow:
|
|
return FoldExponentiation(cx, pn, parser);
|
|
|
|
// Various list nodes not requiring care to minimally fold. Some of
|
|
// these could be further folded/optimized, but we don't make the effort.
|
|
case ParseNodeKind::BitOr:
|
|
case ParseNodeKind::BitXor:
|
|
case ParseNodeKind::BitAnd:
|
|
case ParseNodeKind::StrictEq:
|
|
case ParseNodeKind::Eq:
|
|
case ParseNodeKind::StrictNe:
|
|
case ParseNodeKind::Ne:
|
|
case ParseNodeKind::Lt:
|
|
case ParseNodeKind::Le:
|
|
case ParseNodeKind::Gt:
|
|
case ParseNodeKind::Ge:
|
|
case ParseNodeKind::InstanceOf:
|
|
case ParseNodeKind::In:
|
|
case ParseNodeKind::Comma:
|
|
case ParseNodeKind::Array:
|
|
case ParseNodeKind::Object:
|
|
case ParseNodeKind::StatementList:
|
|
case ParseNodeKind::ClassMethodList:
|
|
case ParseNodeKind::TemplateStringList:
|
|
case ParseNodeKind::Var:
|
|
case ParseNodeKind::Const:
|
|
case ParseNodeKind::Let:
|
|
case ParseNodeKind::ParamsBody:
|
|
case ParseNodeKind::CallSiteObj:
|
|
case ParseNodeKind::ExportSpecList:
|
|
case ParseNodeKind::ImportSpecList:
|
|
return FoldList(cx, pn, parser);
|
|
|
|
case ParseNodeKind::InitialYield:
|
|
MOZ_ASSERT(pn->isArity(PN_UNARY));
|
|
MOZ_ASSERT(pn->pn_kid->isKind(ParseNodeKind::Assign) &&
|
|
pn->pn_kid->pn_left->isKind(ParseNodeKind::Name) &&
|
|
pn->pn_kid->pn_right->isKind(ParseNodeKind::Generator));
|
|
return true;
|
|
|
|
case ParseNodeKind::YieldStar:
|
|
MOZ_ASSERT(pn->isArity(PN_UNARY));
|
|
return Fold(cx, &pn->pn_kid, parser);
|
|
|
|
case ParseNodeKind::Yield:
|
|
case ParseNodeKind::Await:
|
|
MOZ_ASSERT(pn->isArity(PN_UNARY));
|
|
if (!pn->pn_kid)
|
|
return true;
|
|
return Fold(cx, &pn->pn_kid, parser);
|
|
|
|
case ParseNodeKind::Return:
|
|
return FoldReturn(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Try:
|
|
return FoldTry(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Catch:
|
|
return FoldCatch(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Class:
|
|
return FoldClass(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Elem:
|
|
return FoldElement(cx, pnp, parser);
|
|
|
|
case ParseNodeKind::Add:
|
|
return FoldAdd(cx, pnp, parser);
|
|
|
|
case ParseNodeKind::Call:
|
|
case ParseNodeKind::New:
|
|
case ParseNodeKind::SuperCall:
|
|
case ParseNodeKind::TaggedTemplate:
|
|
return FoldCall(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Arguments:
|
|
return FoldArguments(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Switch:
|
|
case ParseNodeKind::Colon:
|
|
case ParseNodeKind::Assign:
|
|
case ParseNodeKind::AddAssign:
|
|
case ParseNodeKind::SubAssign:
|
|
case ParseNodeKind::BitOrAssign:
|
|
case ParseNodeKind::BitAndAssign:
|
|
case ParseNodeKind::BitXorAssign:
|
|
case ParseNodeKind::LshAssign:
|
|
case ParseNodeKind::RshAssign:
|
|
case ParseNodeKind::UrshAssign:
|
|
case ParseNodeKind::DivAssign:
|
|
case ParseNodeKind::ModAssign:
|
|
case ParseNodeKind::MulAssign:
|
|
case ParseNodeKind::PowAssign:
|
|
case ParseNodeKind::Import:
|
|
case ParseNodeKind::ExportFrom:
|
|
case ParseNodeKind::Shorthand:
|
|
case ParseNodeKind::For:
|
|
case ParseNodeKind::ClassMethod:
|
|
case ParseNodeKind::ImportSpec:
|
|
case ParseNodeKind::ExportSpec:
|
|
case ParseNodeKind::SetThis:
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
return Fold(cx, &pn->pn_left, parser) &&
|
|
Fold(cx, &pn->pn_right, parser);
|
|
|
|
case ParseNodeKind::NewTarget:
|
|
case ParseNodeKind::ImportMeta:
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
MOZ_ASSERT(pn->pn_left->isKind(ParseNodeKind::PosHolder));
|
|
MOZ_ASSERT(pn->pn_right->isKind(ParseNodeKind::PosHolder));
|
|
return true;
|
|
|
|
case ParseNodeKind::ClassNames:
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
if (ParseNode*& outerBinding = pn->pn_left) {
|
|
if (!Fold(cx, &outerBinding, parser))
|
|
return false;
|
|
}
|
|
return Fold(cx, &pn->pn_right, parser);
|
|
|
|
case ParseNodeKind::DoWhile:
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
return Fold(cx, &pn->pn_left, parser) &&
|
|
FoldCondition(cx, &pn->pn_right, parser);
|
|
|
|
case ParseNodeKind::While:
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
return FoldCondition(cx, &pn->pn_left, parser) &&
|
|
Fold(cx, &pn->pn_right, parser);
|
|
|
|
case ParseNodeKind::Case: {
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
|
|
// pn_left is null for DefaultClauses.
|
|
if (pn->pn_left) {
|
|
if (!Fold(cx, &pn->pn_left, parser))
|
|
return false;
|
|
}
|
|
return Fold(cx, &pn->pn_right, parser);
|
|
}
|
|
|
|
case ParseNodeKind::With:
|
|
MOZ_ASSERT(pn->isArity(PN_BINARY));
|
|
return Fold(cx, &pn->pn_left, parser) &&
|
|
Fold(cx, &pn->pn_right, parser);
|
|
|
|
case ParseNodeKind::ForIn:
|
|
case ParseNodeKind::ForOf:
|
|
return FoldForInOrOf(cx, pn, parser);
|
|
|
|
case ParseNodeKind::ForHead:
|
|
return FoldForHead(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Label:
|
|
MOZ_ASSERT(pn->isArity(PN_NAME));
|
|
return Fold(cx, &pn->pn_expr, parser);
|
|
|
|
case ParseNodeKind::PropertyName:
|
|
MOZ_CRASH("unreachable, handled by ::Dot");
|
|
|
|
case ParseNodeKind::Dot:
|
|
return FoldDottedProperty(cx, pn, parser);
|
|
|
|
case ParseNodeKind::LexicalScope:
|
|
MOZ_ASSERT(pn->isArity(PN_SCOPE));
|
|
if (!pn->scopeBody())
|
|
return true;
|
|
return Fold(cx, &pn->pn_u.scope.body, parser);
|
|
|
|
case ParseNodeKind::Name:
|
|
return FoldName(cx, pn, parser);
|
|
|
|
case ParseNodeKind::Limit: // invalid sentinel value
|
|
MOZ_CRASH("invalid node kind");
|
|
}
|
|
|
|
MOZ_CRASH("shouldn't reach here");
|
|
return false;
|
|
}
|
|
|
|
bool
|
|
frontend::FoldConstants(JSContext* cx, ParseNode** pnp, PerHandlerParser<FullParseHandler>* parser)
|
|
{
|
|
// Don't constant-fold inside "use asm" code, as this could create a parse
|
|
// tree that doesn't type-check as asm.js.
|
|
if (parser->pc->useAsmOrInsideUseAsm())
|
|
return true;
|
|
|
|
AutoTraceLog traceLog(TraceLoggerForCurrentThread(cx), TraceLogger_BytecodeFoldConstants);
|
|
return Fold(cx, pnp, *parser);
|
|
}
|