/* TODO: stmt_parse -> parse_stmt, etc. */ typedef enum { TYPE_VOID, TYPE_UNKNOWN, TYPE_BUILTIN, TYPE_FN, TYPE_TUPLE, TYPE_ARR, TYPE_PTR } TypeKind; typedef enum { BUILTIN_I8, BUILTIN_I16, BUILTIN_I32, BUILTIN_I64, BUILTIN_U8, BUILTIN_U16, BUILTIN_U32, BUILTIN_U64, BUILTIN_F32, BUILTIN_F64, BUILTIN_TYPE_COUNT } BuiltinType; #define TYPE_FLAG_FLEXIBLE 0x01 #define TYPE_FLAG_RESOLVED 0x02 typedef struct Type { Location where; TypeKind kind; unsigned short flags; union { BuiltinType builtin; struct { Array types; /* [0] = ret_type, [1..] = param_types */ } fn; Array tuple; struct { struct Type *of; union { UInteger n; /* this is NOT set by parse_type; it will be handled by types.c */ struct Expression *n_expr; }; } arr; struct { struct Type *of; } ptr; }; } Type; typedef enum { BLOCK_FN, BLOCK_EXPR } BlockKind; typedef struct Block { BlockKind kind; Location start; Location end; Array stmts; struct Block *parent; struct Expression *ret_expr; /* the return expression of this block, e.g. {foo(); 3} => 3 NULL for no expression. */ } Block; typedef enum { EXPR_LITERAL_FLOAT, EXPR_LITERAL_INT, EXPR_LITERAL_STR, EXPR_IDENT, /* variable or constant */ EXPR_BINARY_OP, EXPR_UNARY_OP, EXPR_FN, EXPR_CALL, EXPR_BLOCK, EXPR_DIRECT } ExprKind; typedef enum { UNARY_MINUS, UNARY_ADDRESS, /* &x */ UNARY_DEREF } UnaryOp; typedef enum { BINARY_SET, /* e.g. x = y */ BINARY_PLUS, BINARY_MINUS, BINARY_MUL, BINARY_DIV, BINARY_COMMA, BINARY_AT_INDEX /* e.g. x[i] */ } BinaryOp; typedef struct { Directive which; Array args; /* of Expression */ } DirectExpr; typedef struct { struct Expression *fn; Array args; /* of Expression */ } CallExpr; #define EXPR_FLAG_FOUND_TYPE 0x01 typedef struct Expression { Location where; ExprKind kind; uint16_t flags; Type type; union { Floating floatl; UInteger intl; StrLiteral strl; struct { UnaryOp op; struct Expression *of; } unary; struct { BinaryOp op; struct Expression *lhs; struct Expression *rhs; } binary; CallExpr call; DirectExpr direct; Identifier ident; struct FnExpr *fn; Block block; }; } Expression; #define DECL_FLAG_ANNOTATES_TYPE 0x01 #define DECL_FLAG_CONST 0x02 #define DECL_FLAG_HAS_EXPR 0x04 #define DECL_FLAG_FOUND_TYPE 0x08 #define DECL_FLAG_ERRORED_ABOUT_SELF_REFERENCE 0x10 /* has there been an error about this decl referencing itself? */ /* OPTIM: Instead of using dynamic arrays, do two passes. */ typedef struct Declaration { Location where; Array idents; Type type; uint16_t flags; Expression expr; struct Value *val; /* only for constant decls. set to NULL here, and to actual value by types.c. */ } Declaration; typedef struct FnExpr { Declaration params; /* declaration of the parameters to this function */ Type ret_type; Block body; } FnExpr; /* an expression such as fn(x: int) int {return 2 * x;} */ typedef enum { STMT_DECL, STMT_EXPR } StatementKind; #define STMT_FLAG_VOIDED_EXPR 0x01 /* the "4;" in fn () { 4; } is a voided expression, but the "4" in fn () int { 4 } is not */ typedef struct { Location where; StatementKind kind; unsigned short flags; union { Declaration decl; Expression expr; }; } Statement; typedef struct { Array stmts; } ParsedFile; typedef struct { Tokenizer *tokr; BlockArr exprs; /* a dynamic array of expressions, so that we don't need to call malloc every time we make an expression */ Block *block; /* which block are we in? NULL = file scope */ } Parser; typedef enum { DECL_END_SEMICOLON, DECL_END_RPAREN } DeclEndType; static bool parse_expr(Parser *p, Expression *e, Token *end); #define PARSE_DECL_ALLOW_CONST_WITH_NO_EXPR 0x01 static bool parse_decl(Parser *p, Declaration *d, DeclEndType ends_with, uint16_t flags); static const char *unary_op_to_str(UnaryOp u) { switch (u) { case UNARY_MINUS: return "-"; case UNARY_ADDRESS: return "&"; case UNARY_DEREF: return "*"; } assert(0); return ""; } static const char *binary_op_to_str(BinaryOp b) { switch (b) { case BINARY_PLUS: return "+"; case BINARY_MINUS: return "-"; case BINARY_MUL: return "*"; case BINARY_DIV: return "/"; case BINARY_SET: return "="; case BINARY_COMMA: return ","; case BINARY_AT_INDEX: return "[]"; } assert(0); return ""; } static bool type_builtin_is_integer(BuiltinType b) { switch (b) { case BUILTIN_I8: case BUILTIN_I16: case BUILTIN_I32: case BUILTIN_I64: case BUILTIN_U8: case BUILTIN_U16: case BUILTIN_U32: case BUILTIN_U64: return true; default: return false; } } static bool type_builtin_is_floating(BuiltinType b) { switch (b) { case BUILTIN_F32: case BUILTIN_F64: return true; default: return false; } } static bool type_builtin_is_numerical(BuiltinType b) { return type_builtin_is_integer(b) || type_builtin_is_floating(b); } /* returns BUILTIN_TYPE_COUNT on failure */ static BuiltinType kw_to_builtin_type(Keyword kw) { switch (kw) { case KW_I8: return BUILTIN_I8; case KW_I16: return BUILTIN_I16; case KW_I32: return BUILTIN_I32; case KW_I64: return BUILTIN_I64; case KW_INT: return BUILTIN_I64; case KW_U8: return BUILTIN_U8; case KW_U16: return BUILTIN_U16; case KW_U32: return BUILTIN_U32; case KW_U64: return BUILTIN_U64; case KW_FLOAT: return BUILTIN_F32; case KW_F32: return BUILTIN_F32; case KW_F64: return BUILTIN_F64; default: return BUILTIN_TYPE_COUNT; } } static Keyword builtin_type_to_kw(BuiltinType t) { switch (t) { case BUILTIN_I8: return KW_I8; case BUILTIN_I16: return KW_I16; case BUILTIN_I32: return KW_I32; case BUILTIN_I64: return KW_I64; case BUILTIN_U8: return KW_U8; case BUILTIN_U16: return KW_U16; case BUILTIN_U32: return KW_U32; case BUILTIN_U64: return KW_U64; case BUILTIN_F32: return KW_F32; case BUILTIN_F64: return KW_F64; case BUILTIN_TYPE_COUNT: break; } assert(0); return KW_COUNT; } /* returns the number of characters written, not including the null character */ static size_t type_to_str_(Type *t, char *buffer, size_t bufsize) { switch (t->kind) { case TYPE_VOID: return str_copy(buffer, bufsize, "void"); case TYPE_UNKNOWN: return str_copy(buffer, bufsize, "???"); case TYPE_BUILTIN: { const char *s = keywords[builtin_type_to_kw(t->builtin)]; return str_copy(buffer, bufsize, s); } case TYPE_FN: { /* number of chars written */ size_t written = str_copy(buffer, bufsize, "fn ("); Type *ret_type = t->fn.types.data; Type *param_types = ret_type + 1; size_t nparams = t->fn.types.len - 1; for (size_t i = 0; i < nparams; i++) { if (i > 0) written += str_copy(buffer + written, bufsize - written, ", "); written += type_to_str_(¶m_types[i], buffer + written, bufsize - written); } written += str_copy(buffer + written, bufsize - written, ")"); if (ret_type->kind != TYPE_VOID) { written += str_copy(buffer + written, bufsize - written, " "); written += type_to_str_(ret_type, buffer + written, bufsize - written); } return written; } break; case TYPE_ARR: { size_t written = str_copy(buffer, bufsize, "["); if (t->flags & TYPE_FLAG_RESOLVED) { snprintf(buffer + written, bufsize - written, UINTEGER_FMT, t->arr.n); written += strlen(buffer + written); } else { written += str_copy(buffer + written, bufsize - written, "N"); } written += str_copy(buffer + written, bufsize - written, "]"); written += type_to_str_(t->arr.of, buffer + written, bufsize - written); return written; } break; case TYPE_TUPLE: { size_t written = str_copy(buffer, bufsize, "("); arr_foreach(&t->tuple, Type, child) { if (child != t->tuple.data) written += str_copy(buffer + written, bufsize - written, ", "); written += type_to_str_(child, buffer + written, bufsize - written); } written += str_copy(buffer + written, bufsize - written, ")"); return written; } case TYPE_PTR: { size_t written = str_copy(buffer, bufsize, "*"); written += type_to_str_(t->ptr.of, buffer + written, bufsize - written); return written; } } assert(0); return 0; } /* return value should be freed by caller */ static char *type_to_str(Type *t) { /* TODO allow types >255 chars */ char *ret = err_malloc(256); type_to_str_(t, ret, 256); return ret; } /* allocate a new expression. */ static Expression *parser_new_expr(Parser *p) { return block_arr_add(&p->exprs); } #define NOT_AN_OP -1 static int op_precedence(Keyword op) { switch (op) { case KW_EQ: return 0; case KW_COMMA: return 5; case KW_PLUS: return 10; case KW_MINUS: return 20; case KW_AMPERSAND: return 25; case KW_ASTERISK: return 30; case KW_SLASH: return 40; default: return NOT_AN_OP; } } /* TODO: check that we check which thing ends it everywhere */ #define EXPR_CAN_END_WITH_COMMA 0x01 /* a comma could end the expression */ static Token *expr_find_end(Parser *p, uint16_t flags, bool *is_vbs) { Tokenizer *t = p->tokr; int paren_level = 0; int brace_level = 0; int square_level = 0; Token *token = t->token; bool could_be_vbs = false; /* could this be a void block statement (whose semicolons can be omitted)? e.g. {x := 5;} */ if (is_vbs) *is_vbs = false; while (1) { if (token->kind == TOKEN_KW) { switch (token->kw) { case KW_COMMA: if ((flags & EXPR_CAN_END_WITH_COMMA) && paren_level == 0 && brace_level == 0 && square_level == 0) return token; break; case KW_LPAREN: paren_level++; break; case KW_RPAREN: paren_level--; if (paren_level < 0) return token; break; case KW_LSQUARE: square_level++; break; case KW_RSQUARE: square_level--; if (square_level < 0) return token; break; case KW_LBRACE: brace_level++; could_be_vbs = true; break; case KW_RBRACE: brace_level--; if (brace_level == 0 && could_be_vbs) { if (is_vbs) *is_vbs = true; return token + 1; /* token afer } is end */ } if (brace_level < 0) return token; break; case KW_SEMICOLON: if (brace_level == 0) return token; could_be_vbs = true; break; default: break; } if (token->kw != KW_RBRACE && token->kw != KW_SEMICOLON && token->kw != KW_LBRACE) could_be_vbs = false; } else { could_be_vbs = false; } if (token->kind == TOKEN_EOF) { if (brace_level > 0) { tokr_err(t, "Opening brace { was never closed."); /* FEATURE: Find out where this is */ } else if (paren_level > 0) { tokr_err(t, "Opening parenthesis ( was never closed."); } else if (square_level > 0) { tokr_err(t, "Opening square bracket [ was never closed."); } else { tokr_err(t, "Could not find end of expression (did you forget a semicolon?)."); /* FEATURE: improve err message */ } t->token = token; /* don't try to continue */ return NULL; } token++; } } static bool parse_type(Parser *p, Type *type) { Tokenizer *t = p->tokr; type->where = t->token->where; type->flags = 0; switch (t->token->kind) { case TOKEN_KW: type->kind = TYPE_BUILTIN; type->builtin = kw_to_builtin_type(t->token->kw); if (type->builtin != BUILTIN_TYPE_COUNT) { t->token++; break; } /* Not a builtin */ switch (t->token->kw) { case KW_FN: { /* function type */ type->kind = TYPE_FN; arr_create(&type->fn.types, sizeof(Type)); t->token++; if (!token_is_kw(t->token, KW_LPAREN)) { tokr_err(t, "Expected ( for function type."); return false; } arr_add(&type->fn.types); /* add return type */ t->token++; if (!token_is_kw(t->token, KW_RPAREN)) { while (1) { Type *param_type = arr_add(&type->fn.types); if (!parse_type(p, param_type)) return false; if (token_is_kw(t->token, KW_RPAREN)) break; if (!token_is_kw(t->token, KW_COMMA)) { tokr_err(t, "Expected , to continue function type parameter list."); return false; } t->token++; /* move past , */ } } t->token++; /* move past ) */ Type *ret_type = type->fn.types.data; /* if there's a symbol that isn't [ or (, that can't be the start of a type */ if (t->token->kind == TOKEN_KW && t->token->kw <= KW_LAST_SYMBOL && t->token->kw != KW_LSQUARE && t->token->kw != KW_LPAREN) { ret_type->kind = TYPE_VOID; ret_type->flags = 0; } else { if (!parse_type(p, ret_type)) return false; } break; } case KW_LSQUARE: { /* array type */ Token *start = t->token; type->kind = TYPE_ARR; t->token++; /* move past [ */ Token *end = expr_find_end(p, 0, NULL); type->arr.n_expr = parser_new_expr(p); if (!parse_expr(p, type->arr.n_expr, end)) return false; t->token = end + 1; /* go past ] */ type->arr.of = err_malloc(sizeof *type->arr.of); /* OPTIM */ if (!parse_type(p, type->arr.of)) return false; type->flags = 0; type->where = start->where; break; } case KW_LPAREN: /* tuple! */ type->kind = TYPE_TUPLE; arr_create(&type->tuple, sizeof(Type)); t->token++; /* move past ( */ while (1) { Type *child = arr_add(&type->tuple); if (!parse_type(p, child)) return false; if (token_is_kw(t->token, KW_RPAREN)) { /* we're done with the tuple */ t->token++; /* move past ) */ break; } if (token_is_kw(t->token, KW_COMMA)) { t->token++; /* move past , */ continue; } else { tokr_err(t, "Expected , to list next tuple type or ) to end tuple type."); return false; } } break; case KW_ASTERISK: /* pointer */ type->kind = TYPE_PTR; type->ptr.of = err_malloc(sizeof *type->ptr.of); /* OPTIM */ t->token++; /* move past * */ if (!parse_type(p, type->ptr.of)) return false; break; default: tokr_err(t, "Unrecognized type."); return false; } break; default: tokr_err(t, "Unrecognized type."); return false; } return true; } static bool parse_stmt(Parser *p, Statement *s); static bool parse_block(Parser *p, Block *b) { Tokenizer *t = p->tokr; Block *prev_block = p->block; b->parent = prev_block; p->block = b; if (!token_is_kw(t->token, KW_LBRACE)) { tokr_err(t, "Expected '{' to open block."); return false; } b->start = t->token->where; t->token++; /* move past { */ arr_create(&b->stmts, sizeof(Statement)); bool ret = true; b->ret_expr = NULL; /* default to no return unless overwritten later */ if (!token_is_kw(t->token, KW_RBRACE)) { /* non-empty block */ while (1) { Statement *stmt = arr_add(&b->stmts); bool success = parse_stmt(p, stmt); if (!success) { ret = false; } if (token_is_kw(t->token, KW_RBRACE)) { if (success && stmt->kind == STMT_EXPR) { if (!(stmt->flags & STMT_FLAG_VOIDED_EXPR)) { b->ret_expr = parser_new_expr(p); *b->ret_expr = stmt->expr; arr_remove_last(&b->stmts); /* only keep this expression in the return value */ } } break; } if (success) { if (stmt->kind == STMT_EXPR && !(stmt->flags & STMT_FLAG_VOIDED_EXPR)) { /* in theory, this should never happen right now */ err_print(stmt->where, "Non-voided expression is not the last statement in a block (you might want to add a ';' to the end of this statement)."); return false; } } if (t->token->kind == TOKEN_EOF) { tokr_err(t, "Expected '}' to close function body."); return false; } } } else { b->ret_expr = NULL; } b->end = t->token->where; t->token++; /* move past } */ p->block = prev_block; return ret; } static bool parse_fn_expr(Parser *p, FnExpr *f) { Tokenizer *t = p->tokr; /* only called when token is fn */ assert(token_is_kw(t->token, KW_FN)); t->token++; if (!token_is_kw(t->token, KW_LPAREN)) { tokr_err(t, "Expected '(' after 'fn'."); return false; } t->token++; parse_decl(p, &f->params, DECL_END_RPAREN, PARSE_DECL_ALLOW_CONST_WITH_NO_EXPR); if (token_is_kw(t->token, KW_LBRACE)) { /* void function */ f->ret_type.kind = TYPE_VOID; f->ret_type.flags = 0; } else { if (!parse_type(p, &f->ret_type)) { return false; } } f->body.kind = BLOCK_FN; return parse_block(p, &f->body); } /* parses, e.g. "(3, 5, foo)" */ static bool parse_args(Parser *p, Array *args) { Tokenizer *t = p->tokr; Token *start = t->token; assert(token_is_kw(start, KW_LPAREN)); arr_create(args, sizeof(Expression)); t->token++; /* move past ( */ if (!token_is_kw(t->token, KW_RPAREN)) { /* non-empty arg list */ while (1) { if (t->token->kind == TOKEN_EOF) { tokr_err(t, "Expected argument list to continue."); info_print(start->where, "This is where the argument list starts."); return false; } Expression *arg = arr_add(args); if (!parse_expr(p, arg, expr_find_end(p, EXPR_CAN_END_WITH_COMMA, NULL))) { return false; } if (token_is_kw(t->token, KW_RPAREN)) break; assert(token_is_kw(t->token, KW_COMMA)); t->token++; /* move past , */ } } t->token++; /* move past ) */ return true; } static void fprint_expr(FILE *out, Expression *e); static bool parse_expr(Parser *p, Expression *e, Token *end) { Tokenizer *t = p->tokr; e->flags = 0; if (end == NULL) return false; e->where = t->token->where; if (end <= t->token) { tokr_err(t, "Empty expression."); return false; } if (end - t->token == 1) { /* 1-token expression */ switch (t->token->kind) { case TOKEN_NUM_LITERAL: { NumLiteral *num = &t->token->num; switch (num->kind) { case NUM_LITERAL_FLOAT: e->kind = EXPR_LITERAL_FLOAT; e->floatl = num->floatval; break; case NUM_LITERAL_INT: e->kind = EXPR_LITERAL_INT; e->intl = num->intval; break; } } break; case TOKEN_IDENT: e->kind = EXPR_IDENT; e->ident = t->token->ident; break; case TOKEN_STR_LITERAL: e->kind = EXPR_LITERAL_STR; e->strl = t->token->str; break; default: tokr_err(t, "Unrecognized expression."); return false; } t->token = end; return true; } Token *start = t->token; if (token_is_kw(t->token, KW_FN)) { /* this is a function */ e->kind = EXPR_FN; e->fn = err_malloc(sizeof *e->fn); if (!parse_fn_expr(p, e->fn)) return false; if (t->token != end) { tokr_err(t, "Direct function calling in an expression is not supported yet.\nYou can wrap the function in parentheses."); /* TODO */ return false; } return true; } /* Find the lowest-precedence operator not in parentheses/braces/square brackets */ int paren_level = 0; int brace_level = 0; int square_level = 0; int lowest_precedence = NOT_AN_OP; /* e.g. (5+3) */ bool entirely_within_parentheses = token_is_kw(t->token, KW_LPAREN); Token *lowest_precedence_op; for (Token *token = t->token; token < end; token++) { if (token->kind == TOKEN_KW) { switch (token->kw) { case KW_LPAREN: paren_level++; break; case KW_RPAREN: paren_level--; if (paren_level == 0 && token != end - 1) entirely_within_parentheses = false; if (paren_level < 0) { t->token = token; tokr_err(t, "Excessive closing )."); t->token = end + 1; return false; } break; case KW_LBRACE: brace_level++; break; case KW_RBRACE: brace_level--; if (brace_level < 0) { t->token = token; tokr_err(t, "Excessive closing }."); return false; } break; case KW_LSQUARE: square_level++; break; case KW_RSQUARE: square_level--; if (square_level < 0) { tokr_err(t, "Excessive closing ]."); return false; } break; default: { /* OPTIM: use individual cases for each op */ if (paren_level == 0 && brace_level == 0 && square_level == 0) { int precedence = op_precedence(token->kw); if (precedence == NOT_AN_OP) break; /* nvm it's not an operator */ if (lowest_precedence == NOT_AN_OP || precedence <= lowest_precedence) { lowest_precedence = precedence; lowest_precedence_op = token; } } } break; } } } if (paren_level > 0) { t->token = start; tokr_err(t, "Too many opening parentheses (."); return false; } if (brace_level > 0) { t->token = start; tokr_err(t, "Too many opening braces {."); return false; } if (square_level > 0) { t->token = start; tokr_err(t, "Too many opening square brackets [."); return false; } if (entirely_within_parentheses) { t->token++; /* move past opening ( */ Token *new_end = end - 1; /* parse to ending ) */ if (!parse_expr(p, e, new_end)) return false; t->token++; /* move past closing ) */ return true; } if (lowest_precedence == NOT_AN_OP) { /* function calls, array accesses, etc. */ /* try a function call or array access */ Token *token = t->token; /* currently unnecessary: paren_level = square_level = 0; */ /* can't call at start, e.g. in (fn() {})(), it is not the empty function "" being called with fn() {} as an argument */ if (token_is_kw(t->token, KW_LPAREN)) { paren_level++; token++; } /* which opening bracket starts the call/array access */ Token *opening_bracket = NULL; for (; token < end; token++) { if (token->kind == TOKEN_KW) { switch (token->kw) { case KW_LPAREN: if (square_level == 0 && paren_level == 0 && brace_level == 0 && token != t->tokens.data && token[-1].kind != TOKEN_DIRECT /* don't include directives */) opening_bracket = token; /* maybe this left parenthesis opens the function call */ paren_level++; break; case KW_LSQUARE: if (square_level == 0 && paren_level == 0 && brace_level == 0) opening_bracket = token; /* (array access) */ square_level++; break; case KW_RPAREN: paren_level--; break; case KW_RSQUARE: square_level--; break; case KW_LBRACE: brace_level++; break; case KW_RBRACE: brace_level--; break; default: break; } } else if (token->kind == TOKEN_EOF) { if (paren_level > 0) { tokr_err(t, "Unmatched ( parenthesis."); return false; } if (square_level > 0) { tokr_err(t, "Unmatched [ square bracket."); return false; } break; } } if (opening_bracket) { switch (opening_bracket->kw) { case KW_LPAREN: { /* it's a function call! */ e->kind = EXPR_CALL; e->call.fn = parser_new_expr(p); if (!parse_expr(p, e->call.fn, opening_bracket)) { /* parse up to ( as function */ return false; } t->token = opening_bracket; return parse_args(p, &e->call.args); } case KW_LSQUARE: { /* it's an array access */ e->kind = EXPR_BINARY_OP; e->binary.op = BINARY_AT_INDEX; e->binary.lhs = parser_new_expr(p); e->binary.rhs = parser_new_expr(p); /* parse array */ if (!parse_expr(p, e->binary.lhs, opening_bracket)) return false; /* parse index */ t->token = opening_bracket + 1; Token *index_end = expr_find_end(p, 0, NULL); if (!parse_expr(p, e->binary.rhs, index_end)) return false; t->token++; /* move past ] */ return true; } default: assert(0); return false; } } if (t->token->kind == TOKEN_DIRECT) { /* it's a directive */ e->kind = EXPR_DIRECT; e->direct.which = t->token->direct; if (token_is_kw(&t->token[1], KW_LPAREN)) { /* has args (but maybe it's just "#foo()") */ t->token++; /* move to ( */ return parse_args(p, &e->direct.args); } else { /* no args */ arr_create(&e->direct.args, sizeof(Expression)); t->token++; return true; } } if (token_is_kw(t->token, KW_LBRACE)) { /* it's a block */ e->kind = EXPR_BLOCK; e->block.kind = BLOCK_EXPR; if (!parse_block(p, &e->block)) return false; if (t->token != end) { tokr_err(t, "Expression continues after end of block."); /* TODO: improve this err message */ return false; } return true; } tokr_err(t, "Not implemented yet."); t->token = end + 1; return false; } /* Check if his is a unary op not a binary one (e.g. +-3 => +(-3), not (+)-(3)). */ while (lowest_precedence_op != t->token && lowest_precedence_op[-1].kind == TOKEN_KW && op_precedence(lowest_precedence_op[-1].kw) != NOT_AN_OP) { lowest_precedence_op--; } if (lowest_precedence_op == t->token) { /* Unary */ UnaryOp op; bool is_unary; switch (lowest_precedence_op->kw) { case KW_PLUS: /* unary + is ignored entirely */ t->token++; /* re-parse this expression without + */ return parse_expr(p, e, end); case KW_MINUS: is_unary = true; op = UNARY_MINUS; break; case KW_AMPERSAND: is_unary = true; op = UNARY_ADDRESS; break; case KW_ASTERISK: is_unary = true; op = UNARY_DEREF; break; default: is_unary = false; break; } if (!is_unary) { tokr_err(t, "%s is not a unary operator.", keywords[lowest_precedence_op->kw]); return false; } e->unary.op = op; e->kind = EXPR_UNARY_OP; t->token++; Expression *of = parser_new_expr(p); e->unary.of = of; return parse_expr(p, of, end); } BinaryOp op; switch (lowest_precedence_op->kw) { case KW_PLUS: op = BINARY_PLUS; break; case KW_MINUS: op = BINARY_MINUS; break; case KW_EQ: op = BINARY_SET; break; case KW_COMMA: op = BINARY_COMMA; break; case KW_ASTERISK: op = BINARY_MUL; break; case KW_SLASH: op = BINARY_DIV; break; default: assert(0); return false; } e->binary.op = op; e->kind = EXPR_BINARY_OP; Expression *lhs = parser_new_expr(p); e->binary.lhs = lhs; if (!parse_expr(p, lhs, lowest_precedence_op)) { return false; } Expression *rhs = parser_new_expr(p); t->token = lowest_precedence_op + 1; e->binary.rhs = rhs; if (!parse_expr(p, rhs, end)) { return false; } return true; } /* parses x : int, y : float; ^^this^^ then recursively calls itself to parse the rest NOTE: this function actually parses all types in the declaration, but it just calls itself to do that. */ static inline bool ends_decl(Token *t, DeclEndType ends_with) { return (token_is_kw(t, KW_SEMICOLON) && ends_with == DECL_END_SEMICOLON) || (token_is_kw(t, KW_RPAREN) && ends_with == DECL_END_RPAREN); } static bool parse_decl(Parser *p, Declaration *d, DeclEndType ends_with, uint16_t flags) { d->val = NULL; d->type.kind = TYPE_VOID; d->where = p->tokr->token->where; arr_create(&d->idents, sizeof(Identifier)); Tokenizer *t = p->tokr; d->flags = 0; if (ends_decl(t->token, ends_with)) { t->token++; return true; } /* OPTIM: Maybe don't use a dynamic array or use parser allocator. */ size_t n_idents_with_this_type = 1; bool ret = true; for (size_t type_no = 0; ; type_no++) { /* parse a single type in this decl */ while (1) { Identifier *ident = arr_add(&d->idents); if (t->token->kind != TOKEN_IDENT) { tokr_err(t, "Cannot declare non-identifier (that's a %s).", token_kind_to_str(t->token->kind)); /* TODO(eventually): an */ ret = false; break; } *ident = t->token->ident; t->token++; if (token_is_kw(t->token, KW_COMMA)) { t->token++; n_idents_with_this_type++; continue; } if (token_is_kw(t->token, KW_COLON)) { if (d->flags & DECL_FLAG_CONST) { tokr_err(t, "Either everything or nothing must be constant in a declaration."); ret = false; break; } t->token++; break; } if (token_is_kw(t->token, KW_AT)) { if (type_no == 0) { d->flags |= DECL_FLAG_CONST; } else if (!(d->flags & DECL_FLAG_CONST)) { tokr_err(t, "Either everything or nothing must be constant in a declaration."); ret = false; break; } t->token++; break; } tokr_err(t, "Expected ',' to continue listing variables or ':' / '@' to indicate type."); ret = false; break; } if (!ret) break; if (token_is_kw(t->token, KW_SEMICOLON) || token_is_kw(t->token, KW_RPAREN)) { /* e.g. foo :; */ tokr_err(t, "Cannot infer type without expression."); ret = false; break; } bool annotates_type = !token_is_kw(t->token, KW_EQ) && !token_is_kw(t->token, KW_COMMA); if (d->type.kind != TYPE_VOID /* multiple types in one declaration */ && (!!(d->flags & DECL_FLAG_ANNOTATES_TYPE)) != annotates_type) { /* annotation on one decl but not the other */ /* e.g. x: int, y := 3, 5;*/ tokr_err(t, "You must specify either all types or no types in a single declaration."); ret = false; break; } if (annotates_type) { d->flags |= DECL_FLAG_ANNOTATES_TYPE; Type type; if (!parse_type(p, &type)) { ret = false; break; } if (n_idents_with_this_type == 1 && d->type.kind == TYPE_VOID) { d->type = type; } else if (d->type.kind == TYPE_TUPLE) { /* add to tuple */ for (size_t i = 0; i < n_idents_with_this_type; i++) { *(Type*)arr_add(&d->type.tuple) = type; } } else { /* construct tuple */ Array tup_arr; arr_create(&tup_arr, sizeof(Type)); if (d->type.kind != TYPE_VOID) { *(Type*)arr_add(&tup_arr) = d->type; /* add current type */ } d->type.flags = 0; d->type.kind = TYPE_TUPLE; d->type.tuple = tup_arr; for (size_t i = 0; i < n_idents_with_this_type; i++) { *(Type*)arr_add(&d->type.tuple) = type; } } } if (token_is_kw(t->token, KW_COMMA)) { /* next type in declaration */ t->token++; /* move past , */ continue; } /* OPTIM: switch t->token->kw ? */ if (token_is_kw(t->token, KW_EQ)) { t->token++; d->flags |= DECL_FLAG_HAS_EXPR; Token *end = expr_find_end(p, 0, NULL); if (!end || !token_is_kw(end, KW_SEMICOLON)) { tokr_err(t, "Expected ';' at end of declaration."); ret = false; break; } if (!parse_expr(p, &d->expr, end)) { t->token = end + 1; /* move past ; */ ret = false; break; } if (ends_decl(t->token, ends_with)) { t->token++; break; } tokr_err(t, "Expected '%c' at end of declaration.", ends_with == DECL_END_SEMICOLON ? ';' : ')'); ret = false; } else if (ends_decl(t->token, ends_with)) { t->token++; } else { tokr_err(t, "Expected '%c' or '=' at end of delaration.", ends_with == DECL_END_SEMICOLON ? ';' : ')'); ret = false; } break; } if ((d->flags & DECL_FLAG_CONST) && !(d->flags & DECL_FLAG_HAS_EXPR) && !(flags & PARSE_DECL_ALLOW_CONST_WITH_NO_EXPR)) { t->token--; /* disallowed constant without an expression, e.g. x @ int; */ tokr_err(t, "You must have an expression at the end of this constant declaration."); ret = false; } if (!ret) { /* move past end of decl */ while (t->token->kind != TOKEN_EOF && !token_is_kw(t->token, KW_SEMICOLON)) { t->token++; } if (token_is_kw(t->token, KW_SEMICOLON)) { t->token++; /* move past ; */ } } return ret; } static bool parse_stmt(Parser *p, Statement *s) { Tokenizer *t = p->tokr; s->flags = 0; if (t->token->kind == TOKEN_EOF) tokr_err(t, "Expected statement."); s->where = t->token->where; /* TODO: statements such as 3, 5; will not work. */ if (token_is_kw(t->token + 1, KW_COLON) || token_is_kw(t->token + 1, KW_COMMA) || token_is_kw(t->token + 1, KW_AT)) { s->kind = STMT_DECL; if (!parse_decl(p, &s->decl, DECL_END_SEMICOLON, 0)) { return false; } return true; } else { s->kind = STMT_EXPR; bool is_vbs; Token *end = expr_find_end(p, 0, &is_vbs); if (!end) { tokr_err(t, "No semicolon found at end of statement."); while (t->token->kind != TOKEN_EOF) t->token++; /* move to end of file */ return false; } if (is_vbs || token_is_kw(end, KW_SEMICOLON)) { s->flags |= STMT_FLAG_VOIDED_EXPR; } bool success = parse_expr(p, &s->expr, end); /* go past end of expr regardless of whether successful or not */ if (token_is_kw(end, KW_SEMICOLON)) t->token = end + 1; /* skip ; */ else t->token = end; return success; } } static void parser_from_tokenizer(Parser *p, Tokenizer *t) { p->tokr = t; p->block = NULL; block_arr_create(&p->exprs, 10, sizeof(Expression)); /* block size = 1024 */ } static bool parse_file(Parser *p, ParsedFile *f) { Tokenizer *t = p->tokr; arr_create(&f->stmts, sizeof(Statement)); bool ret = true; while (t->token->kind != TOKEN_EOF) { Statement *stmt = arr_add(&f->stmts); if (!parse_stmt(p, stmt)) ret = false; if (token_is_kw(t->token, KW_RBRACE)) { tokr_err(t, "} without a matching {."); return false; } } return ret; } #define PARSE_PRINT_LOCATION(l) //fprintf(out, "[l%lu]", (unsigned long)(l).line); static void fprint_expr(FILE *out, Expression *e); static void fprint_stmt(FILE *out, Statement *s); static void fprint_decl(FILE *out, Declaration *d); static void fprint_type(FILE *out, Type *t) { PARSE_PRINT_LOCATION(t->where); fprintf(out, "%s", type_to_str(t)); } static void fprint_block(FILE *out, Block *b) { fprintf(out, "{\n"); arr_foreach(&b->stmts, Statement, stmt) { fprint_stmt(out, stmt); } fprintf(out, "}"); if (b->ret_expr) { fprintf(out, " returns "); fprint_expr(out, b->ret_expr); } } static void fprint_fn_expr(FILE *out, FnExpr *f) { fprintf(out, "fn ("); fprint_decl(out, &f->params); fprintf(out, ") "); fprint_type(out, &f->ret_type); fprintf(out, " "); fprint_block(out, &f->body); } static void fprint_args(FILE *out, Array *args) { fprintf(out, "("); arr_foreach(args, Expression, arg) { if (arg != args->data) fprintf(out, ", "); fprint_expr(out, arg); } fprintf(out, ")"); } static void fprint_expr(FILE *out, Expression *e) { PARSE_PRINT_LOCATION(e->where); switch (e->kind) { case EXPR_LITERAL_INT: fprintf(out, "%lld", (long long)e->intl); break; case EXPR_LITERAL_FLOAT: fprintf(out, "%f", (double)e->floatl); break; case EXPR_LITERAL_STR: fprintf(out, "\"%s\"", e->strl.str); break; case EXPR_IDENT: fprint_ident(out, e->ident); break; case EXPR_BINARY_OP: fprintf(out, "%s", binary_op_to_str(e->binary.op)); fprintf(out, "("); fprint_expr(out, e->binary.lhs); fprintf(out, ","); fprint_expr(out, e->binary.rhs); fprintf(out, ")"); break; case EXPR_UNARY_OP: fprintf(out, "%s", unary_op_to_str(e->unary.op)); fprintf(out, "("); fprint_expr(out, e->unary.of); fprintf(out, ")"); break; case EXPR_FN: fprint_fn_expr(out, e->fn); break; case EXPR_CALL: fprint_expr(out, e->call.fn); fprint_args(out, &e->call.args); break; case EXPR_BLOCK: fprint_block(out, &e->block); break; case EXPR_DIRECT: fprintf(out, "#"); fprintf(out, "%s", directives[e->direct.which]); fprint_args(out, &e->direct.args); break; } fprintf(out, ":"); fprint_type(out, &e->type); } static void fprint_decl(FILE *out, Declaration *d) { PARSE_PRINT_LOCATION(d->where); arr_foreach(&d->idents, Identifier, ident) { if (ident != d->idents.data) fprintf(out, ", "); fprint_ident(out, *ident); } if (d->flags & DECL_FLAG_CONST) { fprintf(out, "[const]"); } fprintf(out, ":"); if ((d->flags & DECL_FLAG_FOUND_TYPE) || (d->flags & DECL_FLAG_ANNOTATES_TYPE)) { fprint_type(out, &d->type); } if (d->flags & DECL_FLAG_HAS_EXPR) { fprintf(out, "="); fprint_expr(out, &d->expr); } } static void fprint_stmt(FILE *out, Statement *s) { PARSE_PRINT_LOCATION(s->where); if (s->flags & STMT_FLAG_VOIDED_EXPR) fprintf(out, "(void)"); switch (s->kind) { case STMT_DECL: fprint_decl(out, &s->decl); fprintf(out, ";\n"); break; case STMT_EXPR: fprint_expr(out, &s->expr); fprintf(out, ";\n"); break; } } static void fprint_parsed_file(FILE *out, ParsedFile *f) { arr_foreach(&f->stmts, Statement, stmt) { fprint_stmt(out, stmt); } } /* TODO: Freeing parser (remember to free args) */