/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- * vim: set ts=8 sts=2 et sw=2 tw=80: * 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/. */ #ifndef frontend_ParseNodeVisitor_h #define frontend_ParseNodeVisitor_h #include "mozilla/Assertions.h" #include "frontend/ParseNode.h" #include "js/friend/StackLimits.h" // js::AutoCheckRecursionLimit namespace js { class FrontendContext; namespace frontend { /** * Utility class for walking a JS AST. * * Simple usage: * * class HowTrueVisitor : public ParseNodeVisitor { * public: * bool visitTrueExpr(BooleanLiteral* pn) { * std::cout << "How true.\n"; * return true; * } * bool visitClassDecl(ClassNode* pn) { * // The base-class implementation of each visit method * // simply visits the node's children. So the subclass * // gets to decide whether to descend into a subtree * // and can do things either before or after: * std::cout << "How classy.\n"; * return ParseNodeVisitor::visitClassDecl(pn); * } * }; * * HowTrueVisitor v; * v.visit(programRootNode); // walks the entire tree * * A ParseNodeVisitor can modify nodes, but it can't replace the current node * with a different one; for that, use a RewritingParseNodeVisitor. * * Note that the Curiously Recurring Template Pattern is used for performance, * as it eliminates the need for virtual method calls. Some rough testing shows * about a 12% speedup in the FoldConstants.cpp pass. * https://en.wikipedia.org/wiki/Curiously_recurring_template_pattern */ template class ParseNodeVisitor { public: FrontendContext* fc_; explicit ParseNodeVisitor(FrontendContext* fc) : fc_(fc) {} [[nodiscard]] bool visit(ParseNode* pn) { AutoCheckRecursionLimit recursion(fc_); if (!recursion.check(fc_)) { return false; } switch (pn->getKind()) { #define VISIT_CASE(KIND, TYPE) \ case ParseNodeKind::KIND: \ return static_cast(this)->visit##KIND(&pn->as()); FOR_EACH_PARSE_NODE_KIND(VISIT_CASE) #undef VISIT_CASE default: MOZ_CRASH("invalid node kind"); } } // using static_cast here allows plain visit() to be overridden. #define VISIT_METHOD(KIND, TYPE) \ [[nodiscard]] bool visit##KIND(TYPE* pn) { /* NOLINT */ \ return pn->accept(*static_cast(this)); \ } FOR_EACH_PARSE_NODE_KIND(VISIT_METHOD) #undef VISIT_METHOD }; /* * Utility class for walking a JS AST that allows for replacing nodes. * * The API is the same as ParseNodeVisitor, except that visit methods take an * argument of type `ParseNode*&`, a reference to the field in the parent node * that points down to the node being visited. Methods can replace the current * node by assigning to this reference. * * All visit methods take a `ParseNode*&` rather than more specific types like * `BinaryNode*&`, to allow replacing the current node with one of a different * type. Constant folding makes use of this. */ template class RewritingParseNodeVisitor { public: FrontendContext* fc_; explicit RewritingParseNodeVisitor(FrontendContext* fc) : fc_(fc) {} [[nodiscard]] bool visit(ParseNode*& pn) { AutoCheckRecursionLimit recursion(fc_); if (!recursion.check(fc_)) { return false; } switch (pn->getKind()) { #define VISIT_CASE(KIND, _type) \ case ParseNodeKind::KIND: \ return static_cast(this)->visit##KIND(pn); FOR_EACH_PARSE_NODE_KIND(VISIT_CASE) #undef VISIT_CASE default: MOZ_CRASH("invalid node kind"); } } // using static_cast here allows plain visit() to be overridden. #define VISIT_METHOD(KIND, TYPE) \ [[nodiscard]] bool visit##KIND(ParseNode*& pn) { \ MOZ_ASSERT(pn->is(), \ "Node of kind " #KIND " was not of type " #TYPE); \ return pn->as().accept(*static_cast(this)); \ } FOR_EACH_PARSE_NODE_KIND(VISIT_METHOD) #undef VISIT_METHOD }; } // namespace frontend } // namespace js #endif // frontend_ParseNodeVisitor_h