// Part of the Carbon Language project, under the Apache License v2.0 with LLVM // Exceptions. See /LICENSE for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception #ifndef CARBON_TOOLCHAIN_SEMANTICS_SEMANTICS_CONTEXT_H_ #define CARBON_TOOLCHAIN_SEMANTICS_SEMANTICS_CONTEXT_H_ #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/DenseSet.h" #include "llvm/ADT/FoldingSet.h" #include "llvm/ADT/SmallVector.h" #include "toolchain/parser/parse_tree.h" #include "toolchain/semantics/semantics_declaration_name_stack.h" #include "toolchain/semantics/semantics_ir.h" #include "toolchain/semantics/semantics_node.h" #include "toolchain/semantics/semantics_node_block_stack.h" #include "toolchain/semantics/semantics_node_stack.h" namespace Carbon { // Context and shared functionality for semantics handlers. class SemanticsContext { public: // Stores references for work. explicit SemanticsContext(const TokenizedBuffer& tokens, DiagnosticEmitter& emitter, const ParseTree& parse_tree, SemanticsIR& semantics, llvm::raw_ostream* vlog_stream); // Marks an implementation TODO. Always returns false. auto TODO(ParseTree::Node parse_node, std::string label) -> bool; // Runs verification that the processing cleanly finished. auto VerifyOnFinish() -> void; // Adds a node to the current block, returning the produced ID. auto AddNode(SemanticsNode node) -> SemanticsNodeId; // Adds a node to the given block, returning the produced ID. auto AddNodeToBlock(SemanticsNodeBlockId block, SemanticsNode node) -> SemanticsNodeId; // Pushes a parse tree node onto the stack, storing the SemanticsNode as the // result. auto AddNodeAndPush(ParseTree::Node parse_node, SemanticsNode node) -> void; // Adds a name to name lookup. Prints a diagnostic for name conflicts. auto AddNameToLookup(ParseTree::Node name_node, SemanticsStringId name_id, SemanticsNodeId target_id) -> void; // Performs name lookup in a specified scope, returning the referenced node. // If scope_id is invalid, uses the current contextual scope. auto LookupName(ParseTree::Node parse_node, SemanticsStringId name_id, SemanticsNameScopeId scope_id, bool print_diagnostics) -> SemanticsNodeId; // Prints a diagnostic for a duplicate name. auto DiagnoseDuplicateName(ParseTree::Node parse_node, SemanticsNodeId prev_def_id) -> void; // Prints a diagnostic for a missing name. auto DiagnoseNameNotFound(ParseTree::Node parse_node, SemanticsStringId name_id) -> void; // Pushes a new scope onto scope_stack_. auto PushScope() -> void; // Pops the top scope from scope_stack_, cleaning up names from name_lookup_. auto PopScope() -> void; // Adds a `Branch` node branching to a new node block, and returns the ID of // the new block. All paths to the branch target must go through the current // block, though not necessarily through this branch. auto AddDominatedBlockAndBranch(ParseTree::Node parse_node) -> SemanticsNodeBlockId; // Adds a `Branch` node branching to a new node block with a value, and // returns the ID of the new block. All paths to the branch target must go // through the current block. auto AddDominatedBlockAndBranchWithArg(ParseTree::Node parse_node, SemanticsNodeId arg_id) -> SemanticsNodeBlockId; // Adds a `BranchIf` node branching to a new node block, and returns the ID // of the new block. All paths to the branch target must go through the // current block. auto AddDominatedBlockAndBranchIf(ParseTree::Node parse_node, SemanticsNodeId cond_id) -> SemanticsNodeBlockId; // Adds branches from the given list of blocks to a new block, for // reconvergence of control flow, and pushes the new block onto the node // block stack. auto AddConvergenceBlockAndPush( ParseTree::Node parse_node, std::initializer_list blocks) -> void; // Adds branches from the given list of blocks and values to a new block, for // reconvergence of control flow with a result value, and pushes the new // block onto the node block stack. Returns a node referring to the result // value. auto AddConvergenceBlockWithArgAndPush( ParseTree::Node parse_node, std::initializer_list> blocks_and_args) -> SemanticsNodeId; // Add the current code block to the enclosing function. auto AddCurrentCodeBlockToFunction() -> void; // Returns whether the current position in the current block is reachable. auto is_current_position_reachable() -> bool; // Runs ImplicitAsImpl for a set of arguments and parameters. // // This will eventually need to support checking against multiple possible // overloads, multiple of which may be possible but not "best". While this can // currently be done by calling twice, toggling `apply_implicit_as`, in the // future we may want to remember the right implicit conversions to do for // valid cases in order to efficiently handle generics. auto ImplicitAsForArgs( SemanticsNodeBlockId arg_refs_id, ParseTree::Node param_parse_node, SemanticsNodeBlockId param_refs_id, DiagnosticEmitter::DiagnosticBuilder* diagnostic) -> bool; // Runs ImplicitAsImpl for a situation where a cast is required, returning the // updated `value_id`. Prints a diagnostic and returns an Error if // unsupported. auto ImplicitAsRequired(ParseTree::Node parse_node, SemanticsNodeId value_id, SemanticsTypeId as_type_id) -> SemanticsNodeId; // Runs ImplicitAsRequired for a conversion to `bool`. auto ImplicitAsBool(ParseTree::Node parse_node, SemanticsNodeId value_id) -> SemanticsNodeId; // Canonicalizes a type which is tracked as a single node. // TODO: This should eventually return a type ID. auto CanonicalizeType(SemanticsNodeId node_id) -> SemanticsTypeId; // Handles canonicalization of struct types. This may create a new struct type // when it has a new structure, or reference an existing struct type when it // duplicates a prior type. // // Individual struct type fields aren't canonicalized because they may have // name conflicts or other diagnostics during creation, which can use the // parse node. auto CanonicalizeStructType(ParseTree::Node parse_node, SemanticsNodeBlockId refs_id) -> SemanticsTypeId; // Handles canonicalization of tuple types. This may create a new tuple type // if the `type_ids` doesn't match an existing tuple type. auto CanonicalizeTupleType(ParseTree::Node parse_node, llvm::SmallVector&& type_ids) -> SemanticsTypeId; // Returns a pointer type whose pointee type is `pointee_type_id`. auto GetPointerType(ParseTree::Node parse_node, SemanticsTypeId pointee_type_id) -> SemanticsTypeId; // Converts an expression for use as a type. // TODO: This should eventually return a type ID. auto ExpressionAsType(ParseTree::Node parse_node, SemanticsNodeId value_id) -> SemanticsTypeId { auto node = semantics_ir_->GetNode(value_id); if (node.kind() == SemanticsNodeKind::StubReference) { value_id = node.GetAsStubReference(); CARBON_CHECK(semantics_ir_->GetNode(value_id).kind() != SemanticsNodeKind::StubReference) << "Stub reference should not point to another stub reference"; } return CanonicalizeType( ImplicitAsRequired(parse_node, value_id, SemanticsTypeId::TypeType)); } // Removes any top-level `const` qualifiers from a type. auto GetUnqualifiedType(SemanticsTypeId type_id) -> SemanticsTypeId; // Starts handling parameters or arguments. auto ParamOrArgStart() -> void; // On a comma, pushes the entry. On return, the top of node_stack_ will be // start_kind. auto ParamOrArgComma(bool for_args) -> void; // Detects whether there's an entry to push. On return, the top of // node_stack_ will be start_kind, and the caller should do type-specific // processing. Returns refs_id. auto ParamOrArgEnd(bool for_args, ParseNodeKind start_kind) -> SemanticsNodeBlockId; // Saves a parameter from the top block in node_stack_ to the top block in // params_or_args_stack_. If for_args, adds a StubReference of the previous // node's result to the IR. // // This should only be called by other ParamOrArg functions, not directly. auto ParamOrArgSave(bool for_args) -> void; // Prints information for a stack dump. auto PrintForStackDump(llvm::raw_ostream& output) const -> void; auto tokens() -> const TokenizedBuffer& { return *tokens_; } auto emitter() -> DiagnosticEmitter& { return *emitter_; } auto parse_tree() -> const ParseTree& { return *parse_tree_; } auto semantics_ir() -> SemanticsIR& { return *semantics_ir_; } auto node_stack() -> SemanticsNodeStack& { return node_stack_; } auto node_block_stack() -> SemanticsNodeBlockStack& { return node_block_stack_; } auto args_type_info_stack() -> SemanticsNodeBlockStack& { return args_type_info_stack_; } auto return_scope_stack() -> llvm::SmallVector& { return return_scope_stack_; } auto declaration_name_stack() -> SemanticsDeclarationNameStack& { return declaration_name_stack_; } private: // For CanImplicitAs, the detected conversion to apply. enum ImplicitAsKind { // Incompatible types. Incompatible, // No conversion required. Identical, // ImplicitAs is required. Compatible, }; // A FoldingSet node for a type. class TypeNode : public llvm::FastFoldingSetNode { public: explicit TypeNode(const llvm::FoldingSetNodeID& node_id, SemanticsTypeId type_id) : llvm::FastFoldingSetNode(node_id), type_id_(type_id) {} auto type_id() -> SemanticsTypeId { return type_id_; } private: SemanticsTypeId type_id_; }; // An entry in scope_stack_. struct ScopeStackEntry { // Names which are registered with name_lookup_, and will need to be // deregistered when the scope ends. llvm::DenseSet names; // TODO: This likely needs to track things which need to be destructed. }; // Runs ImplicitAs behavior to convert `value` to `as_type`, returning the // result type. The result will be the node to use to replace `value`. // // If `output_value_id` is null, then this only checks if the conversion is // possible. // // If `output_value_id` is not null, then it will be set if there is a need to // cast. auto ImplicitAsImpl(SemanticsNodeId value_id, SemanticsTypeId as_type_id, SemanticsNodeId* output_value_id) -> ImplicitAsKind; // Forms a canonical type ID for a type. This function is given two // callbacks: // // `profile_type(canonical_id)` is called to build a fingerprint for this // type. The ID should be distinct for all distinct type values with the same // `kind`. // // `make_node()` is called to obtain a `SemanticsNodeId` that describes the // type. It is only called if the type does not already exist, so can be used // to lazily build the `SemanticsNode`. `make_node()` is not permitted to // directly or indirectly canonicalize any types. auto CanonicalizeTypeImpl( SemanticsNodeKind kind, llvm::function_ref profile_type, llvm::function_ref make_node) -> SemanticsTypeId; // Forms a canonical type ID for a type. If the type is new, adds the node to // the current block. auto CanonicalizeTypeAndAddNodeIfNew(SemanticsNode node) -> SemanticsTypeId; auto current_scope() -> ScopeStackEntry& { return scope_stack_.back(); } // Tokens for getting data on literals. const TokenizedBuffer* tokens_; // Handles diagnostics. DiagnosticEmitter* emitter_; // The file's parse tree. const ParseTree* parse_tree_; // The SemanticsIR being added to. SemanticsIR* semantics_ir_; // Whether to print verbose output. llvm::raw_ostream* vlog_stream_; // The stack during Build. Will contain file-level parse nodes on return. SemanticsNodeStack node_stack_; // The stack of node blocks being used for general IR generation. SemanticsNodeBlockStack node_block_stack_; // The stack of node blocks being used for per-element tracking of nodes in // parameter and argument node blocks. Versus node_block_stack_, an element // will have 1 or more nodes in blocks in node_block_stack_, but only ever 1 // node in blocks here. SemanticsNodeBlockStack params_or_args_stack_; // The stack of node blocks being used for type information while processing // arguments. This is used in parallel with params_or_args_stack_. It's // currently only used for struct literals, where we need to track names // for a type separate from the literal arguments. SemanticsNodeBlockStack args_type_info_stack_; // A stack of return scopes; i.e., targets for `return`. Inside a function, // this will be a FunctionDeclaration. llvm::SmallVector return_scope_stack_; // A stack for scope context. llvm::SmallVector scope_stack_; // The stack used for qualified declaration name construction. SemanticsDeclarationNameStack declaration_name_stack_; // Maps identifiers to name lookup results. Values are a stack of name lookup // results in the ancestor scopes. This offers constant-time lookup of names, // regardless of how many scopes exist between the name declaration and // reference. // // Names which no longer have lookup results are erased. llvm::DenseMap> name_lookup_; // Cache of the mapping from nodes to types, to avoid recomputing the folding // set ID. llvm::DenseMap canonical_types_; // Tracks the canonical representation of types that have been defined. llvm::FoldingSet canonical_type_nodes_; // Storage for the nodes in canonical_type_nodes_. This stores in pointers so // that FoldingSet can have stable pointers. llvm::SmallVector> type_node_storage_; }; // Parse node handlers. Returns false for unrecoverable errors. #define CARBON_PARSE_NODE_KIND(Name) \ auto SemanticsHandle##Name(SemanticsContext& context, \ ParseTree::Node parse_node) \ ->bool; #include "toolchain/parser/parse_node_kind.def" } // namespace Carbon #endif // CARBON_TOOLCHAIN_SEMANTICS_SEMANTICS_CONTEXT_H_