#include "../include/ast.h" #include "../include/parser.h" #include "../include/compiler.h" #include "../include/std/log.h" #include "../include/std/hash.h" #include #include #include #define throw(result) do { throw_debug_msg; longjmp(context->env, (result)); } while(0) #define throw_if_error(result) \ do { \ int return_if_result; \ return_if_result = (result); \ if((return_if_result) != 0) \ throw(return_if_result); \ } while(0) static void variable_type_resolve(VariableType *self, AstCompilerContext *context, AstResolvedType *resolved_type); static void ast_resolve(Ast *self, AstCompilerContext *context); static void scope_named_object_init(ScopeNamedObject *self) { self->type = NAMED_OBJECT_NONE; self->value.data = NULL; self->resolve_data = NULL; } /* TODO: Remove these? */ #if 0 static bool scope_named_object_equals(ScopeNamedObject *self, ScopeNamedObject *other) { /* This should be fine, without checking ScopeNamedObject type, since they are only equal if the types are equal as well */ return self->value.lhs_expr == other->value.lhs_expr; } static BufferView scope_named_object_get_type_name(ScopeNamedObject *self) { BufferView result; switch(self->type) { case NAMED_OBJECT_NONE: result = create_buffer_view("", 0); break; case NAMED_OBJECT_LHS_EXPR: result = variable_type_get_name(&self->value.lhs_expr->type); break; case NAMED_OBJECT_FUNC_PARAM: result = variable_type_get_name(&self->value.func_param->type); break; } return result; } #endif static BufferView ast_resolved_type_get_name(AstResolvedType *self) { BufferView result; switch(self->type) { case RESOLVED_TYPE_NONE: result = create_buffer_view("", 0); break; case RESOLVED_TYPE_LHS_EXPR: result = self->value.lhs_expr->var_name; break; case RESOLVED_TYPE_FUNC_SIG: { if(!self->value.func_sig->func_decl || !self->value.func_sig->func_decl->lhs_expr) { /* TODO: Use function signature string from the source file, which will also help with error reference. Currently this would point to an invalid location in the source file */ result = create_buffer_view("fn()", 0); break; } result = self->value.func_sig->func_decl->lhs_expr->var_name; break; } } return result; } static void ast_resolved_type_init(AstResolvedType *self) { self->value.data = NULL; self->type = RESOLVED_TYPE_NONE; } static bool ast_resolved_type_equals(AstResolvedType *self, AstResolvedType *other) { if(self->type != other->type) return bool_false; switch(self->type) { case RESOLVED_TYPE_NONE: return bool_true; case RESOLVED_TYPE_LHS_EXPR: return self->value.lhs_expr == other->value.lhs_expr; case RESOLVED_TYPE_FUNC_SIG: return function_signature_equals(self->value.func_sig, other->value.func_sig); } return bool_false; } static AstResolvedType lhs_expr_get_resolved_type(LhsExpr *self, AstCompilerContext *context) { AstResolvedType result; variable_type_resolve(&self->type, context, &result); if(result.type != RESOLVED_TYPE_NONE) return result; assert(self->rhs_expr); ast_resolve(self->rhs_expr, context); return self->rhs_expr->resolve_data.type; } /* TODO: Detect recursive dependency? is it even possible for function parameters? (it would normally be a recursive dependency on the function declaration) */ static void function_parameter_resolve(FunctionParameter *self, AstCompilerContext *context) { if(self->resolve_data.status == AST_RESOLVED) return; self->resolve_data.status = AST_RESOLVING; variable_type_resolve(&self->type, context, &self->resolve_data.type); self->resolve_data.status = AST_RESOLVED; } static AstResolvedType scope_named_object_get_resolved_type(ScopeNamedObject *self, AstCompilerContext *context) { AstResolvedType result; switch(self->type) { case NAMED_OBJECT_NONE: /* Type not resolved */ assert(bool_false); ast_resolved_type_init(&result); return result; case NAMED_OBJECT_LHS_EXPR: result = lhs_expr_get_resolved_type(self->value.lhs_expr, context); break; case NAMED_OBJECT_FUNC_PARAM: function_parameter_resolve(self->value.func_param, context); result = self->value.func_param->resolve_data.type; break; } return result; } static void resolve_data_init(AstResolveData *self) { self->status = AST_NOT_RESOLVED; ast_resolved_type_init(&self->type); self->ir_reg = 0; } int ast_create(ArenaAllocator *allocator, void *value, AstType type, Ast **result) { return_if_error(arena_allocator_alloc(allocator, sizeof(Ast), (void**)result)); (*result)->value.data = value; (*result)->type = type; resolve_data_init(&(*result)->resolve_data); (*result)->parser = NULL; return 0; } static bool ast_is_decl(Ast *self) { /* TODO: Add more types as they are introduced */ return self->type == AST_FUNCTION_DECL || self->type == AST_STRUCT_DECL; } BufferView ast_get_name(Ast *self) { BufferView name; switch(self->type) { case AST_FUNCTION_DECL: case AST_STRUCT_DECL: case AST_IMPORT: case AST_STRING: case AST_BINOP: case AST_IF_STATEMENT: case AST_WHILE_STATEMENT: case AST_RETURN: name = create_buffer_view_null(); break; case AST_NUMBER: name = self->value.number->code_ref; break; case AST_BOOL: name = self->value.bool->code_ref; break; case AST_LHS: name = self->value.lhs_expr->var_name; break; case AST_ASSIGN: name = ast_get_name(self->value.assign_expr->lhs_expr); break; case AST_FUNCTION_CALL: name = self->value.func_call->func.name; break; case AST_VARIABLE: name = self->value.variable->name; break; case AST_STRUCT_FIELD: name = self->value.struct_field->name; break; } return name; } static BufferView ast_get_code_reference(Ast *self) { return ast_get_name(self); } int function_signature_init(FunctionSignature *self, ArenaAllocator *allocator) { self->resolved = bool_false; self->func_decl = NULL; return_if_error(buffer_init(&self->parameters, allocator)); return buffer_init(&self->return_types, allocator); } static FunctionParameter* function_signature_get_parameter_by_name(FunctionSignature *self, BufferView name) { FunctionParameter *param, *param_end; param = buffer_begin(&self->parameters); param_end = buffer_end(&self->parameters); for(; param != param_end; ++param) { if(buffer_view_equals(&name, ¶m->name)) return param; } return NULL; } int function_signature_add_parameter(FunctionSignature *self, const FunctionParameter *new_param) { const FunctionParameter *existing_param = function_signature_get_parameter_by_name(self, new_param->name); if(existing_param) return AST_ERR_DEF_DUP; return buffer_append(&self->parameters, new_param, sizeof(FunctionParameter)); } int function_signature_add_return_type(FunctionSignature *self, const VariableType *var_type) { FunctionReturnType return_type; return_type.type = *var_type; ast_resolved_type_init(&return_type.resolved_type); return buffer_append(&self->return_types, &return_type, sizeof(return_type)); } static CHECK_RESULT bool function_parameter_equals(FunctionParameter *self, FunctionParameter *other) { /* It's fine if the name of the parameter is different. Only the type matters */ return ast_resolved_type_equals(&self->resolve_data.type, &other->resolve_data.type); } static CHECK_RESULT bool function_parameters_equals(Buffer *func_params, Buffer *other_func_params) { FunctionParameter *func_param, *func_param_end; FunctionParameter *other_func_param, *other_func_param_end; func_param = buffer_begin(func_params); func_param_end = buffer_end(func_params); other_func_param = buffer_begin(other_func_params); other_func_param_end = buffer_end(other_func_params); /* Different number of arguments */ if(func_param_end - func_param != other_func_param_end - other_func_param) return bool_false; for(; func_param != func_param_end; ++func_param, ++other_func_param) { if(!function_parameter_equals(func_param, other_func_param)) return bool_false; } return bool_true; } static CHECK_RESULT bool function_return_type_equals(FunctionReturnType *self, FunctionReturnType *other) { return ast_resolved_type_equals(&self->resolved_type, &other->resolved_type); } static CHECK_RESULT bool function_return_types_equals(Buffer *func_return_types, Buffer *other_func_return_types) { FunctionReturnType *func_return_type, *func_return_type_end; FunctionReturnType *other_func_return_type, *other_func_return_type_end; func_return_type = buffer_begin(func_return_types); func_return_type_end = buffer_end(func_return_types); other_func_return_type = buffer_begin(other_func_return_types); other_func_return_type_end = buffer_end(other_func_return_types); /* Different number of arguments */ if(func_return_type_end - func_return_type != other_func_return_type_end - other_func_return_type) return bool_false; for(; func_return_type != func_return_type_end; ++func_return_type, ++other_func_return_type) { if(!function_return_type_equals(func_return_type, other_func_return_type)) return bool_false; } return bool_true; } bool function_signature_equals(FunctionSignature *self, FunctionSignature *other) { if(!function_parameters_equals(&self->parameters, &other->parameters)) return bool_false; return function_return_types_equals(&self->return_types, &other->return_types); } void function_parameter_init(FunctionParameter *self) { self->name = create_buffer_view_null(); self->type.type = VARIABLE_TYPE_NONE; self->type.value.variable = NULL; resolve_data_init(&self->resolve_data); } int funcdecl_init(FunctionDecl *self, FunctionSignature *signature, Scope *parent, ArenaAllocator *allocator) { self->lhs_expr = NULL; self->signature = signature; self->ir_func_index = 0; return scope_init(&self->body, parent, allocator); } int funccall_init(FunctionCall *self, BufferView name, ArenaAllocator *allocator) { variable_init(&self->func, name); return buffer_init(&self->args, allocator); } int structdecl_init(StructDecl *self, Scope *parent, ArenaAllocator *allocator) { self->fields_num_pointers = 0; self->fields_fixed_size_bytes = 0; self->is_signed = bool_false; return scope_init(&self->body, parent, allocator); } int structdecl_add_field(StructDecl *self, StructField *field, ArenaAllocator *allocator) { Ast *body_obj; return_if_error(ast_create(allocator, field, AST_STRUCT_FIELD, &body_obj)); return scope_add_child(&self->body, body_obj); } LhsExpr* structdecl_get_field_by_name(StructDecl *self, BufferView field_name) { Ast* result; if(!hash_map_get(&self->body.named_objects, field_name, &result)) return NULL; return result->value.lhs_expr; } void structfield_init(StructField *self, BufferView name, VariableType *type) { self->name = name; self->type = *type; } void lhsexpr_init(LhsExpr *self, DeclFlag decl_flag, BufferView var_name) { assert(!((decl_flag & DECL_FLAG_EXTERN) && (decl_flag & DECL_FLAG_EXPORT)) && "Expression cant be both extern and export"); self->decl_flags = decl_flag; self->type.type = VARIABLE_TYPE_NONE; self->type.value.variable = NULL; self->var_name = var_name; self->rhs_expr = NULL; self->extern_index = 0; } void assignmentexpr_init(AssignmentExpr *self, Ast *lhs_expr, Ast *rhs_expr) { self->lhs_expr = lhs_expr; self->rhs_expr = rhs_expr; } void import_init(Import *self, BufferView path) { self->path = path; self->file_scope = NULL; } int string_init(String *self, BufferView str) { /* TODO: Convert special characters. For example \n should be converted to binary newline etc */ self->str = str; return 0; } void number_init(Number *self, AmalNumber *value, BufferView code_ref) { self->value = *value; self->code_ref = code_ref; } void ast_bool_init(AstBool *self, bool value, BufferView code_ref) { self->value = value; self->code_ref = code_ref; } void variable_init(Variable *self, BufferView name) { self->name = name; scope_named_object_init(&self->resolved_var); } void binop_init(Binop *self) { self->lhs = NULL; self->rhs = NULL; self->type = BINOP_ADD; self->grouped = bool_false; } int if_statement_init(IfStatement *self, Scope *parent, ArenaAllocator *allocator) { self->condition = NULL; self->else_if_stmt = NULL; return scope_init(&self->body, parent, allocator); } int else_if_statement_init(ElseIfStatement *self, Scope *parent, ArenaAllocator *allocator) { self->condition = NULL; self->next_else_if_stmt = NULL; return scope_init(&self->body, parent, allocator); } int while_statement_init(WhileStatement *self, Scope *parent, ArenaAllocator *allocator) { self->condition = NULL; return scope_init(&self->body, parent, allocator); } void return_expr_init(ReturnExpr *self, Ast *rhs_expr) { self->rhs_expr = rhs_expr; } int scope_init(Scope *self, Scope *parent, ArenaAllocator *allocator) { return_if_error(buffer_init(&self->ast_objects, allocator)); return_if_error(hash_map_init(&self->named_objects, allocator, sizeof(Ast*), hash_map_compare_string, amal_hash_string)); self->parent = parent; self->function_signature = NULL; return 0; } int file_scope_reference_init(FileScopeReference *self, BufferView canonical_path, ArenaAllocator *allocator) { char null_terminator; null_terminator = '\0'; self->parser = NULL; return_if_error(buffer_init(&self->canonical_path, allocator)); return_if_error(buffer_append(&self->canonical_path, canonical_path.data, canonical_path.size)); return_if_error(buffer_append(&self->canonical_path, &null_terminator, 1)); /* To exclude null-terminator character from size but not from data */ self->canonical_path.size -= 1; return 0; } int scope_add_child(Scope *self, Ast *child) { Ast *existing_child; bool child_already_exists; if(child->type == AST_LHS) { BufferView var_name; var_name = child->value.lhs_expr->var_name; assert(var_name.data); child_already_exists = hash_map_get(&self->named_objects, var_name, &existing_child); if(child_already_exists) return AST_ERR_DEF_DUP; cleanup_if_error(hash_map_insert(&self->named_objects, var_name, &child)); } cleanup_if_error(buffer_append(&self->ast_objects, &child, sizeof(Ast*))); return 0; cleanup: return AST_ERR; } void scope_resolve(Scope *self, AstCompilerContext *context) { Ast **ast = buffer_begin(&self->ast_objects); Ast **ast_end = buffer_end(&self->ast_objects); Scope *prev_scope = context->scope; context->scope = self; for(; ast != ast_end; ++ast) { ast_resolve(*ast, context); } context->scope = prev_scope; } static void compiler_print_error(amal_compiler *compiler, const char *ref, const char *fmt, ...) { Tokenizer *tokenizer; va_list args; va_start(args, fmt); tokenizer = amal_compiler_find_tokenizer_by_code_reference(compiler, ref); if(!tokenizer) { amal_log_error("Failed to find tokenizer for code reference %p. Is it an invalid reference?", ref ? ref : "(null)"); vfprintf(stderr, fmt, args); fputc('\n', stderr); va_end(args); return; } tokenizer_print_error_args(tokenizer, tokenizer_get_code_reference_index(tokenizer, ref), fmt, args); va_end(args); } static void __scope_get_resolved_variable(Scope *self, Scope *start, AstCompilerContext *context, BufferView name, ScopeNamedObject *result) { Ast *ast_result; bool exists; Scope *prev_scope; assert(self); exists = hash_map_get(&self->named_objects, name, &ast_result); if(!exists) { if(self->function_signature) { FunctionParameter *func_param = function_signature_get_parameter_by_name(self->function_signature, name); if(func_param) { prev_scope = context->scope; context->scope = self; function_parameter_resolve(func_param, context); context->scope = prev_scope; result->type = NAMED_OBJECT_FUNC_PARAM; result->value.func_param = func_param; result->resolve_data = &func_param->resolve_data; return; } /* TODO: Remove this when closures can capture variables */ assert(self->parent == &context->parser->struct_decl.body); } if(self->parent) { __scope_get_resolved_variable(self->parent, start, context, name, result); return; } compiler_print_error(context->compiler, name.data, "Undefined reference to variable \"%.*s\"", name.size, name.data); throw(AST_ERR); } /* Need to change scope here because we are changing the visible scope and the ast object may be in another scope than the current resolving ast. */ prev_scope = context->scope; context->scope = self; ast_resolve(ast_result, context); context->scope = prev_scope; assert(ast_result->type == AST_LHS); result->type = NAMED_OBJECT_LHS_EXPR; result->value.lhs_expr = ast_result->value.lhs_expr; result->resolve_data = &ast_result->resolve_data; } static void scope_get_resolved_variable(Scope *self, AstCompilerContext *context, BufferView name, ScopeNamedObject *result) { __scope_get_resolved_variable(self, self, context, name, result); } static void function_return_type_resolve(FunctionReturnType *self, AstCompilerContext *context) { variable_type_resolve(&self->type, context, &self->resolved_type); } static void function_signature_resolve(FunctionSignature *self, AstCompilerContext *context) { if(self->resolved) return; { FunctionParameter *param, *param_end; param = buffer_begin(&self->parameters); param_end = buffer_end(&self->parameters); for(; param != param_end; ++param) { function_parameter_resolve(param, context); } } { FunctionReturnType *return_type, *return_type_end; return_type = buffer_begin(&self->return_types); return_type_end = buffer_end(&self->return_types); for(; return_type != return_type_end; ++return_type) { function_return_type_resolve(return_type, context); } } self->resolved = bool_true; } static void variable_resolve(Variable *self, AstCompilerContext *context, AstResolvedType *resolved_type) { if(self->resolved_var.type == NAMED_OBJECT_NONE) scope_get_resolved_variable(context->scope, context, self->name, &self->resolved_var); *resolved_type = scope_named_object_get_resolved_type(&self->resolved_var, context); } void variable_type_resolve(VariableType *self, AstCompilerContext *context, AstResolvedType *resolved_type) { switch(self->type) { case VARIABLE_TYPE_NONE: ast_resolved_type_init(resolved_type); return; case VARIABLE_TYPE_VARIABLE: variable_resolve(self->value.variable, context, resolved_type); break; case VARIABLE_TYPE_SIGNATURE: function_signature_resolve(self->value.signature, context); resolved_type->type = RESOLVED_TYPE_FUNC_SIG; resolved_type->value.func_sig = self->value.signature; break; } } static void lhsexpr_resolve_rhs(LhsExpr *self, AstCompilerContext *context, AstResolvedType *result) { ast_resolve(self->rhs_expr, context); if(ast_is_decl(self->rhs_expr)) { result->type = RESOLVED_TYPE_LHS_EXPR; result->value.lhs_expr = self; /* Structs resolved type becomes the lhs while functions resolved type becomes the function signature they own */ if (self->rhs_expr->type == AST_STRUCT_DECL) self->rhs_expr->resolve_data.type = *result; } else { *result = self->rhs_expr->resolve_data.type; } } static void lhsexpr_resolve(Ast *ast, AstCompilerContext *context) { LhsExpr *self; assert(ast->type == AST_LHS); self = ast->value.lhs_expr; variable_type_resolve(&self->type, context, &ast->resolve_data.type); /* TODO: When parameters and return types are implemented, AST_RESOLVE_END should be set after the parameters and return types have been resolved as recursive function calls should be allowed but recursive function calls still require parameters and return types to be known. */ if(self->rhs_expr) { AstResolvedType rhs_resolve_type; if(self->rhs_expr->type == AST_FUNCTION_DECL) { /* The function declaration itself always resolves the signature, but we also do it here because we want to have the signature solved before setting the lhs expr as solved. Also function signatures can exist without lhs expr (anonymous function). */ function_signature_resolve(self->rhs_expr->value.func_decl->signature, context); ast->resolve_data.status = AST_RESOLVED; /* If rhs is a function declaration then there is no need to wait until it has been resolved before setting the type. We still need to continue after this, so rhs can be resolved. */ if(ast->resolve_data.type.type == RESOLVED_TYPE_NONE) { ast->resolve_data.type.type = RESOLVED_TYPE_FUNC_SIG; ast->resolve_data.type.value.func_sig = self->rhs_expr->value.func_decl->signature; } } lhsexpr_resolve_rhs(self, context, &rhs_resolve_type); /* TODO: Add casting */ if(ast->resolve_data.type.type == RESOLVED_TYPE_LHS_EXPR && !ast_resolved_type_equals(&ast->resolve_data.type, &rhs_resolve_type)) { /* TODO: Instead of using self->var_name, use type name. This cant be done right now because type can be function signature. */ compiler_print_error(context->compiler, self->var_name.data, "Variable type and variable assignment type (right-hand side) do not match"); throw(AST_ERR); } ast->resolve_data.type = rhs_resolve_type; } } static LhsExpr* binop_get_lhs_expr(Binop *self) { if(self->rhs) { if(self->rhs->type == AST_LHS) return self->rhs->value.lhs_expr; else if(self->rhs->type == AST_BINOP) return binop_get_lhs_expr(self->rhs->value.binop); } else { if(self->lhs->type == AST_LHS) return self->lhs->value.lhs_expr; else if(self->lhs->type == AST_BINOP) return binop_get_lhs_expr(self->lhs->value.binop); } return NULL; } static void assignmentexpr_resolve(Ast *ast, AstCompilerContext *context) { AssignmentExpr *self; bool is_lhs_const; is_lhs_const = bool_false; assert(ast->type == AST_ASSIGN); self = ast->value.assign_expr; ast_resolve(self->lhs_expr, context); ast_resolve(self->rhs_expr, context); if(self->lhs_expr->type == AST_VARIABLE) { /* TODO: Allow non-const function param */ const ScopeNamedObject *resolved_var = &self->lhs_expr->value.variable->resolved_var; if(resolved_var->type == NAMED_OBJECT_FUNC_PARAM || LHS_EXPR_IS_CONST(resolved_var->value.lhs_expr)) is_lhs_const = bool_true; } else if(self->lhs_expr->type == AST_BINOP) { LhsExpr *lhs_expr = binop_get_lhs_expr(self->lhs_expr->value.binop); check(lhs_expr); is_lhs_const = LHS_EXPR_IS_CONST(lhs_expr); } /* This also covers extern variables, since extern variables are always const */ /* TODO: var.field type expressions should also be checked */ if(is_lhs_const) { compiler_print_error(context->compiler, ast_get_code_reference(self->lhs_expr).data, "Can't assign to a const variable"); throw(AST_ERR); } /* TODO: Add casting */ if(!ast_resolved_type_equals(&self->lhs_expr->resolve_data.type, &self->rhs_expr->resolve_data.type)) { BufferView rhs_type_name = ast_resolved_type_get_name(&self->rhs_expr->resolve_data.type); BufferView lhs_type_name = ast_resolved_type_get_name(&self->lhs_expr->resolve_data.type); /* TODO: Instead of using self->var_name, use type name. This cant be done right now because type can be function signature. */ compiler_print_error(context->compiler, ast_get_code_reference(self->lhs_expr).data, "Can't cast data of type %.*s to type %.*s", rhs_type_name.size, rhs_type_name.data, lhs_type_name.size, lhs_type_name.data); throw(AST_ERR); } ast->resolve_data.type = self->lhs_expr->resolve_data.type; } static void import_resolve(Ast *ast, AstCompilerContext *context) { Import *self; assert(ast->type == AST_IMPORT); (void)context; self = ast->value.import; ast->resolve_data.type.type = RESOLVED_TYPE_LHS_EXPR; ast->resolve_data.type.value.lhs_expr = &self->file_scope->file_scope_ref->parser->file_decl; } static Scope* lhsexpr_get_scope(LhsExpr *self) { AstValue value = self->rhs_expr->value; switch(self->rhs_expr->type) { case AST_FUNCTION_DECL: return &value.func_decl->body; case AST_STRUCT_DECL: return &value.struct_decl->body; case AST_IMPORT: /* *import_index = 1 + value.import->file_scope->import_index;*/ assert(bool_false); return &value.import->file_scope->file_scope_ref->parser->struct_decl.body; default: break; } assert(bool_false && "Expected lhsexpr_get_scope to only be called for non-extern function declaration, struct declaration and import"); return NULL; } static Scope* ast_resolved_type_get_scope(AstResolvedType *self) { switch(self->type) { case RESOLVED_TYPE_NONE: assert(bool_false && "Expected ast_resolved_type_get_scope to only be called for a resolved object"); return NULL; case RESOLVED_TYPE_LHS_EXPR: return lhsexpr_get_scope(self->value.lhs_expr); case RESOLVED_TYPE_FUNC_SIG: assert(self->value.func_sig->func_decl); return &self->value.func_sig->func_decl->body; } return NULL; } static void funcdecl_resolve(Ast *self, AstCompilerContext *context) { FunctionDecl *func_decl = self->value.func_decl; function_signature_resolve(func_decl->signature, context); scope_resolve(&func_decl->body, context); self->resolve_data.type.type = RESOLVED_TYPE_FUNC_SIG; self->resolve_data.type.value.func_sig = self->value.func_decl->signature; } static usize min(usize a, usize b) { return a < b ? a : b; } static bool is_c_pointer_compatible(VariableType *self) { return self->variable_type_flags & VARIABLE_TYPE_FLAG_BORROW; } static bool is_arg_str_and_param_c_str(AstResolvedType *arg_type, VariableType *param_type, AstCompilerContext *context) { return arg_type->value.data == &context->compiler->default_types.str->lhs_expr && param_type->type == VARIABLE_TYPE_VARIABLE && param_type->value.variable->resolved_var.value.data == context->compiler->default_types.c_char && is_c_pointer_compatible(param_type); } static bool resolve_data_type_equals(AstResolvedType *self, AstResolvedType *other) { if(self->type != other->type) return bool_false; switch(self->type) { case RESOLVED_TYPE_NONE: case RESOLVED_TYPE_LHS_EXPR: return self->value.data == other->value.data; case RESOLVED_TYPE_FUNC_SIG: return function_signature_equals(self->value.func_sig, other->value.func_sig); } check(bool_false); return bool_false; } static bool function_parameter_is_c_vararg(FunctionParameter *self, AstCompilerContext *context) { amal_default_type *vararg_type = context->compiler->default_types.c_varargs; return self->resolve_data.type.value.data == &vararg_type->lhs_expr; } static bool is_function_arg_compatible_with_parameter(AstResolvedType *arg, FunctionParameter *param, AstCompilerContext *context) { return resolve_data_type_equals(arg, ¶m->resolve_data.type) || is_arg_str_and_param_c_str(arg, ¶m->type, context) || function_parameter_is_c_vararg(param, context); } /* Pointers, isize and usize are returned with size 4, as that is the smallest possible size for them */ static int arithmetic_type_get_size(StructDecl *self) { assert(sizeof(usize) >= 4 && "Make this work when size of pointer is less than 4"); return self->fields_num_pointers * 4 + self->fields_fixed_size_bytes; } static bool is_implicit_cast_possible(AstResolvedType *from, FunctionParameter *to, AstCompilerContext *context) { if(from->type == RESOLVED_TYPE_LHS_EXPR && to->resolve_data.type.type == RESOLVED_TYPE_LHS_EXPR) { LhsExpr *from_lhs_expr = from->value.lhs_expr; LhsExpr *to_lhs_expr = to->resolve_data.type.value.lhs_expr; assert(to_lhs_expr->rhs_expr && to_lhs_expr->rhs_expr->type == AST_STRUCT_DECL); /* TODO: Optimize, dont use is_arithmetic_type */ if(is_arithmetic_type(from_lhs_expr, context->compiler) && is_arithmetic_type(to_lhs_expr, context->compiler)) { StructDecl *from_struct_type = from_lhs_expr->rhs_expr->value.struct_decl; StructDecl *to_struct_type = to_lhs_expr->rhs_expr->value.struct_decl; assert(from_lhs_expr->rhs_expr->type == AST_STRUCT_DECL); /* TODO: Also allow implicit cast if @from variable is const and if the const value can be cast to @to without data loss */ /* TODO: Allow cast between signed<->unsigned if the value is known at compile-time and if it can be cast without data loss */ return arithmetic_type_get_size(from_struct_type) <= arithmetic_type_get_size(to_struct_type) && from_struct_type->is_signed == to_struct_type->is_signed; } } return bool_false; } static void funccall_resolve_signature_types(FunctionCall *func_call, FunctionSignature *func_sig, AstCompilerContext *context) { Ast **arg = buffer_begin(&func_call->args); Ast **arg_end = buffer_end(&func_call->args); FunctionParameter *func_param = buffer_begin(&func_sig->parameters); FunctionParameter *func_param_end = buffer_end(&func_sig->parameters); usize num_args = arg_end - arg; usize num_params = func_param_end - func_param; isize num_missing_args = (isize)num_params - (isize)num_args; usize num_check = min(num_args, num_params); usize i = 0; for(; i < num_check; ++i) { if(!is_function_arg_compatible_with_parameter(&(*arg)->resolve_data.type, func_param, context)) { /* TODO: Cast data to larger size? */ if(!is_implicit_cast_possible(&(*arg)->resolve_data.type, func_param, context)) { BufferView arg_name = ast_resolved_type_get_name(&(*arg)->resolve_data.type); BufferView param_name = ast_resolved_type_get_name(&func_param->resolve_data.type); /* TODO: Use arg as tokenizer reference, but the name is generated right now so it doesn't belong to the tokenizer; so the reference cant be used. */ compiler_print_error(context->compiler, func_call->func.name.data, "Can't implicitly cast argument of type \"%.*s\" to parameter of type \"%.*s\"", arg_name.size, arg_name.data, param_name.size, param_name.data); throw(AST_ERR); } } ++arg; ++func_param; } if(num_missing_args > 0) { FunctionParameter *vararg_param = buffer_begin(&func_sig->parameters); bool has_vararg = num_params > 0 && function_parameter_is_c_vararg(&vararg_param[num_params - 1], context); if (has_vararg) num_missing_args -= 1; if(num_missing_args > 0) { compiler_print_error(context->compiler, func_call->func.name.data, "Missing %d argument(s) to closure \"%.*s\"", num_missing_args, func_call->func.name.size, func_call->func.name.data); throw(AST_ERR); } } } static void funccall_resolve(Ast *self, AstCompilerContext *context) { Ast **ast; Ast **ast_end; FunctionSignature *func_sig; FunctionCall *func_call = self->value.func_call; variable_resolve(&func_call->func, context, &self->resolve_data.type); /* Attemping to use call syntax (variable_name ( ) ) with a variable that is not a function */ if(self->resolve_data.type.type != RESOLVED_TYPE_FUNC_SIG) { BufferView callee_code_ref = ast_resolved_type_get_name(&self->resolve_data.type); compiler_print_error(context->compiler, func_call->func.name.data, "\"%.*s\" is not a function. Only functions can be called", func_call->func.name.size, func_call->func.name.data); /* TODO: use tokenizer_print_note, once it has been added */ /* TODO: Print type */ compiler_print_error(context->compiler, callee_code_ref.data, "Type was declared here"); throw(AST_ERR); } { func_sig = self->resolve_data.type.value.func_sig; self->resolve_data.type.type = RESOLVED_TYPE_LHS_EXPR; if(func_sig->return_types.size > 0) { FunctionReturnType *return_type = buffer_begin(&func_sig->return_types); self->resolve_data.type = return_type->resolved_type; } else { self->resolve_data.type.type = RESOLVED_TYPE_LHS_EXPR; self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.void_type->lhs_expr; } } ast = buffer_begin(&func_call->args); ast_end = buffer_end(&func_call->args); for(; ast != ast_end; ++ast) { ast_resolve(*ast, context); } funccall_resolve_signature_types(func_call, func_sig, context); } static TypeSize variable_type_get_byte_size(VariableType *self) { TypeSize type_size; type_size.num_pointers = 0; type_size.fixed_size = 0; switch(self->type) { case VARIABLE_TYPE_NONE: assert(bool_false && "Variable type not resolved!"); break; case VARIABLE_TYPE_VARIABLE: { if(self->variable_type_flags & VARIABLE_TYPE_FLAG_BORROW) type_size.num_pointers = 1; else type_size = resolved_type_get_byte_size(&self->value.variable->resolved_var.resolve_data->type); break; } case VARIABLE_TYPE_SIGNATURE: type_size.num_pointers = 1; break; } return type_size; } static void structdecl_resolve(Ast *self, AstCompilerContext *context) { StructDecl *struct_decl = self->value.struct_decl; Scope *body = &struct_decl->body; scope_resolve(body, context); { /* Sum the size of all the fields into the struct, so the struct can know it's full size without searching for it. TODO: Exclude functions, but not function pointers. */ Ast **ast = buffer_begin(&body->ast_objects); Ast **ast_end = buffer_end(&body->ast_objects); for(; ast != ast_end; ++ast) { StructField *struct_field = (*ast)->value.struct_field; TypeSize type_size = variable_type_get_byte_size(&struct_field->type); struct_decl->fields_num_pointers += type_size.num_pointers; struct_decl->fields_fixed_size_bytes += type_size.fixed_size; } } } static void structfield_resolve(Ast *self, AstCompilerContext *context) { StructField *struct_field = self->value.struct_field; variable_type_resolve(&struct_field->type, context, &self->resolve_data.type); } static bool is_struct_decl(Ast *self) { LhsExpr *resolved_type; if(self->resolve_data.type.type != RESOLVED_TYPE_LHS_EXPR) return bool_false; resolved_type = self->resolve_data.type.value.lhs_expr; assert(self->resolve_data.status == AST_RESOLVED); return resolved_type->rhs_expr && resolved_type->rhs_expr->type == AST_STRUCT_DECL; } static void binop_resolve_dot_access(Ast *ast, AstCompilerContext *context, ScopeNamedObject *rhs_resolved_var) { Binop *self; Scope *lhs_scope; BufferView caller_code_ref; BufferView callee_code_ref; assert(ast->type == AST_BINOP); self = ast->value.binop; if(self->lhs->type != AST_VARIABLE) { /* TODO: Allow field access for numbers and string as well */ compiler_print_error(context->compiler, ast_get_code_reference(self->lhs).data, "Accessing fields is only applicable for variables"); throw(AST_ERR); } lhs_scope = ast_resolved_type_get_scope(&self->lhs->resolve_data.type); scope_get_resolved_variable(lhs_scope, context, self->rhs->value.variable->name, rhs_resolved_var); self->rhs->resolve_data.type = scope_named_object_get_resolved_type(rhs_resolved_var, context); caller_code_ref = ast_get_code_reference(self->rhs); callee_code_ref = ast_resolved_type_get_name(&self->rhs->resolve_data.type); if(!is_struct_decl(self->lhs)) { compiler_print_error(context->compiler, caller_code_ref.data, "Can only access field of structs"); /* TODO: use tokenizer_print_note, once it has been added */ /* TODO: Print type */ compiler_print_error(context->compiler, callee_code_ref.data, "Type was declared here"); throw(AST_ERR); } { bool invalid_dot_access = bool_true; switch(self->rhs->resolve_data.type.type) { case RESOLVED_TYPE_NONE: assert(bool_false); break; case RESOLVED_TYPE_FUNC_SIG: { FunctionSignature *func_sig = self->rhs->resolve_data.type.value.func_sig; if(func_sig->func_decl && func_sig->func_decl->lhs_expr && LHS_EXPR_IS_PUB(func_sig->func_decl->lhs_expr)) invalid_dot_access = bool_false; break; } case RESOLVED_TYPE_LHS_EXPR: invalid_dot_access = !LHS_EXPR_IS_PUB(self->rhs->resolve_data.type.value.lhs_expr); break; } if(invalid_dot_access) { compiler_print_error(context->compiler, caller_code_ref.data, "Can't access non-public field \"%.*s\"", caller_code_ref.size, caller_code_ref.data); /* TODO: use tokenizer_print_note, once it has been added */ /* TODO: Print type */ compiler_print_error(context->compiler, callee_code_ref.data, "Type was declared non-public here"); throw(AST_ERR); } } } static void binop_resolve(Ast *ast, AstCompilerContext *context) { Binop *self; assert(ast->type == AST_BINOP); self = ast->value.binop; ast_resolve(self->lhs, context); if(self->type == BINOP_DOT && (self->rhs->type == AST_VARIABLE || self->rhs->type == AST_FUNCTION_CALL)) { ScopeNamedObject rhs_resolved_var; binop_resolve_dot_access(ast, context, &rhs_resolved_var); /* Only function call has extra data that needs to be resolved (args) */ if(self->rhs->type == AST_FUNCTION_CALL) { /*Scope *prev_scope = context->scope;*/ /*context->scope = ast_resolved_type_get_scope(&self->lhs->resolve_data.type);*/ self->rhs->value.func_call->func.resolved_var = rhs_resolved_var; ast_resolve(self->rhs, context); /*context->scope = prev_scope;*/ } self->rhs->resolve_data.status = AST_RESOLVED; ast->resolve_data.type = self->rhs->resolve_data.type; } else { ast_resolve(self->rhs, context); /* TODO: Convert types that can be safely converted */ assert(self->lhs->resolve_data.type.type != RESOLVED_TYPE_NONE); assert(self->rhs->resolve_data.type.type != RESOLVED_TYPE_NONE); if(!ast_resolved_type_equals(&self->rhs->resolve_data.type, &self->lhs->resolve_data.type)) { /* TODO: For this first error, only print the line without a reference to code. This requires change in tokenizer_print_error to be able to take a line as reference. TODO: Use note for the additional information instead of error. */ BufferView lhs_type_name = ast_resolved_type_get_name(&self->lhs->resolve_data.type); BufferView rhs_type_name = ast_resolved_type_get_name(&self->rhs->resolve_data.type); compiler_print_error(context->compiler, ast_get_code_reference(self->rhs).data, "Can't cast type %.*s to type %.*s", rhs_type_name.size, rhs_type_name.data, lhs_type_name.size, lhs_type_name.data); compiler_print_error(context->compiler, ast_get_code_reference(self->lhs).data, "Left-hand side is of type %.*s", lhs_type_name.size, lhs_type_name.data); compiler_print_error(context->compiler, ast_get_code_reference(self->rhs).data, "Right-hand side is of type %.*s", rhs_type_name.size, rhs_type_name.data); throw(AST_ERR); /* TODO: Optimize this? store arithmetic type in the LhsExpr itself? */ } else if(self->lhs->resolve_data.type.type != RESOLVED_TYPE_LHS_EXPR || !is_arithmetic_type(self->lhs->resolve_data.type.value.lhs_expr, context->compiler)) { /* TODO: Point the error at the binop instead of LHS */ BufferView lhs_type_name = ast_resolved_type_get_name(&self->lhs->resolve_data.type); compiler_print_error(context->compiler, ast_get_code_reference(self->lhs).data, "Arithmetic operation can only be performed with the types i8, u8, i16, u16, i32, u32, i64, u64, isize and usize", lhs_type_name.size, lhs_type_name.data); throw(AST_ERR); } ast->resolve_data.type = self->lhs->resolve_data.type; } } static void else_if_statement_resolve(ElseIfStatement *else_if_stmt, AstCompilerContext *context) { if(else_if_stmt->condition) ast_resolve(else_if_stmt->condition, context); scope_resolve(&else_if_stmt->body, context); if(else_if_stmt->next_else_if_stmt) else_if_statement_resolve(else_if_stmt->next_else_if_stmt, context); } static void if_statement_resolve(IfStatement *if_stmt, AstCompilerContext *context) { assert(if_stmt->condition); ast_resolve(if_stmt->condition, context); scope_resolve(&if_stmt->body, context); if(if_stmt->else_if_stmt) else_if_statement_resolve(if_stmt->else_if_stmt, context); } static void while_statement_resolve(WhileStatement *while_stmt, AstCompilerContext *context) { ast_resolve(while_stmt->condition, context); scope_resolve(&while_stmt->body, context); } static void return_expr_resolve(ReturnExpr *self, AstCompilerContext *context) { ast_resolve(self->rhs_expr, context); } TypeSize resolved_type_get_byte_size(AstResolvedType *self) { TypeSize type_size; type_size.num_pointers = 0; type_size.fixed_size = 0; switch(self->type) { case RESOLVED_TYPE_NONE: assert(bool_false && "Type not resolved!"); break; case RESOLVED_TYPE_LHS_EXPR: { /* Resolved type until rhs is StructDecl or FunctionSignature */ LhsExpr *lhs_expr = self->value.lhs_expr; if(lhs_expr->type.type != VARIABLE_TYPE_NONE) type_size = variable_type_get_byte_size(&lhs_expr->type); else { assert(lhs_expr->rhs_expr); if(lhs_expr->rhs_expr->type == AST_STRUCT_DECL) { StructDecl *struct_decl = lhs_expr->rhs_expr->value.struct_decl; type_size.num_pointers = struct_decl->fields_num_pointers; type_size.fixed_size = struct_decl->fields_fixed_size_bytes; } else { type_size = resolved_type_get_byte_size(&lhs_expr->rhs_expr->resolve_data.type); } } break; } case RESOLVED_TYPE_FUNC_SIG: type_size.num_pointers = 1; break; } return type_size; } static void signed_integer_resolve(Ast *self, AstCompilerContext *context) { Number *number = self->value.number; assert(number->value.type == AMAL_NUMBER_SIGNED_INTEGER); switch(number->value.bits) { case 8: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.i8->lhs_expr; break; case 16: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.i16->lhs_expr; break; case 32: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.i32->lhs_expr; break; case 64: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.i64->lhs_expr; break; default: { compiler_print_error(context->compiler, ast_get_code_reference(self).data, "Currently only 8, 16, 32 and 64 bit integers are supported. TODO: Support arbitrary size integer"); throw(AST_ERR); break; } } } static void unsigned_integer_resolve(Ast *self, AstCompilerContext *context) { Number *number = self->value.number; assert(number->value.type == AMAL_NUMBER_UNSIGNED_INTEGER); switch(number->value.bits) { case 8: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.u8->lhs_expr; break; case 16: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.u16->lhs_expr; break; case 32: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.u32->lhs_expr; break; case 64: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.u64->lhs_expr; break; default: { compiler_print_error(context->compiler, ast_get_code_reference(self).data, "Currently only 8, 16, 32 and 64 bit integers are supported. TODO: Support arbitrary size integer"); throw(AST_ERR); break; } } } static void float_resolve(Ast *self, AstCompilerContext *context) { Number *number = self->value.number; assert(number->value.type == AMAL_NUMBER_FLOAT); switch(number->value.bits) { case 32: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.f32->lhs_expr; break; case 64: self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.f64->lhs_expr; break; default: { compiler_print_error(context->compiler, ast_get_code_reference(self).data, "Currently only 32 and 64 bit floats are supported. TODO: Support arbitrary size float"); throw(AST_ERR); break; } } } static void number_resolve(Ast *self, AstCompilerContext *context) { Number *number = self->value.number; self->resolve_data.type.type = RESOLVED_TYPE_LHS_EXPR; switch(number->value.type) { case AMAL_NUMBER_SIGNED_INTEGER: signed_integer_resolve(self, context); break; case AMAL_NUMBER_UNSIGNED_INTEGER: unsigned_integer_resolve(self, context); break; case AMAL_NUMBER_FLOAT: float_resolve(self, context); break; } } void ast_resolve(Ast *self, AstCompilerContext *context) { assert(self); assert(context->parser); /* TODO: Move these to the types that need checks for recursive dependency (function declaration, struct declaration) For function declaration, it should be marked as resolved when the signature has been resolved instead of the whole function declaration including the body because the body can have function call that calls functions that are resolving or even recursive function call, which should be allowed. */ /* This check is outside lhs_expr mutex for optimization purpose as most times there wont be a race in multiple threads to resolve an AST expression. */ if(self->resolve_data.status == AST_RESOLVED) { return; /* NOTE!!!: There is a data race here and it's fine! A recursive dependency can be missed here in one thread, but it's fine because that can only happen if another thread is also resolving the same ast, in which case that thread will finally report the recursive dependency. Note that this can mean that different files can report the recursive dependency everytime you compile code with recursive dependency. However this can happen even without a data race since files are compile in parallel so one file might resolve the ast with the issue before another file has done it, and it can be different everytime the files are compiled. */ } else if(self->resolve_data.status == AST_RESOLVING && self->parser == context->parser) { compiler_print_error(context->compiler, ast_get_code_reference(self).data, "Found recursive dependency"); throw(AST_ERR); } /* TODO: This could cause a deadlock if two different threads resolve the same expression at the same time and self->parser is overwritten by one of the threads, and then if both of the threads try to resolve another expression at the same time that this expression depends on, then the same thread could get ownership of that expression (self->parser would be assigned to that thread) or another thread steals it. Then the other thread that had its self->parser stolen would be stuck in a recursive dependency deadlock since in the above code, an error is only thrown if the parser belongs to the current thread. A possible fix for this may be to add a check above that if any of the other thread has failed, then this thread should fail as well. */ self->resolve_data.status = AST_RESOLVING; self->parser = context->parser; switch(self->type) { case AST_NUMBER: number_resolve(self, context); break; case AST_BOOL: { self->resolve_data.type.type = RESOLVED_TYPE_LHS_EXPR; self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.bool->lhs_expr; break; } case AST_FUNCTION_DECL: funcdecl_resolve(self, context); break; case AST_FUNCTION_CALL: funccall_resolve(self, context); break; case AST_STRUCT_DECL: structdecl_resolve(self, context); break; case AST_STRUCT_FIELD: structfield_resolve(self, context); break; case AST_LHS: lhsexpr_resolve(self, context); break; case AST_ASSIGN: assignmentexpr_resolve(self, context); break; case AST_IMPORT: /* TODO: When @import(...).data syntax is added, implement the resolve for it */ import_resolve(self, context); break; case AST_STRING: /* TODO: Convert special combinations. For example \n to newline */ self->resolve_data.type.type = RESOLVED_TYPE_LHS_EXPR; self->resolve_data.type.value.lhs_expr = &context->compiler->default_types.str->lhs_expr; break; case AST_VARIABLE: variable_resolve(self->value.variable, context, &self->resolve_data.type); break; case AST_BINOP: binop_resolve(self, context); break; case AST_IF_STATEMENT: if_statement_resolve(self->value.if_stmt, context); break; case AST_WHILE_STATEMENT: while_statement_resolve(self->value.while_stmt, context); break; case AST_RETURN: return_expr_resolve(self->value.return_expr, context); break; } /* TODO: See comment at the top of this function */ /*assert(self->resolve_data.type.type != RESOLVED_TYPE_NONE);*/ self->resolve_data.status = AST_RESOLVED; }