proper call typing?

This commit is contained in:
pommicket 2022-02-07 14:52:50 -05:00
parent c2377fa40f
commit 9372a72d43
4 changed files with 93 additions and 8 deletions

View file

@ -222,7 +222,7 @@
; array of n t's: TYPE_ARRAY {n as 8 bytes} t ; array of n t's: TYPE_ARRAY {n as 8 bytes} t
; struct/union: TYPE_STRUCT {8-byte pointer to struct/union data (see structures in main.b)} ; struct/union: TYPE_STRUCT {8-byte pointer to struct/union data (see structures in main.b)}
; note: incomplete structs/unions are replaced with void. ; note: incomplete structs/unions are replaced with void.
; function: TYPE_FUNCTION {arg1 type} {arg2 type} ... {argn type} 0 {return type} ; function: TYPE_FUNCTION {arg1 type} {arg2 type} ... {argn type} 0 {return type} - NB: varargs (...) are ignored
; note that enum types are just treated as ints. ; note that enum types are just treated as ints.
#define TYPE_VOID 1 #define TYPE_VOID 1
#define TYPE_CHAR 3 #define TYPE_CHAR 3

View file

@ -53,6 +53,8 @@ global global_variables
; - the first one is a STATEMENT_LOCAL_DECLARATION (with dat2=dat3=dat4=0), which is only there to set rsp properly because of parameters ; - the first one is a STATEMENT_LOCAL_DECLARATION (with dat2=dat3=dat4=0), which is only there to set rsp properly because of parameters
; - the second one is the function body (a STATEMENT_BLOCK) ; - the second one is the function body (a STATEMENT_BLOCK)
global function_statements global function_statements
; ident list mapping function names to function types (TYPE_FUNCTION {...})
global function_types
; statement_datas[0] = pointer to statement data for block-nesting depth 0 (i.e. function bodies) ; statement_datas[0] = pointer to statement data for block-nesting depth 0 (i.e. function bodies)
; statement_datas[1] = pointer to statement data for block-nesting depth 1 (blocks inside functions) ; statement_datas[1] = pointer to statement data for block-nesting depth 1 (blocks inside functions)
; statement_datas[2] = pointer to statement data for block-nesting depth 2 (blocks inside blocks inside functions) ; statement_datas[2] = pointer to statement data for block-nesting depth 2 (blocks inside blocks inside functions)
@ -232,8 +234,8 @@ function main
enumerators = ident_list_create(4000000) enumerators = ident_list_create(4000000)
structures = ident_list_create(4000000) structures = ident_list_create(4000000)
global_variables = ident_list_create(400000) global_variables = ident_list_create(400000)
function_statements = ident_list_create(400000) function_statements = ident_list_create(800000)
function_types = ident_list_create(800000)
function_stmt_data = malloc(800000) ; should be at least 40 bytes * max # of functions function_stmt_data = malloc(800000) ; should be at least 40 bytes * max # of functions
dat_banned_objmacros = 255 dat_banned_objmacros = 255

View file

@ -1,9 +1,15 @@
/* static int g; */ /* static int g; */
int f(int a, float x[], double y, ...) {
}
float * g() {
}
int main() { int main() {
int a = exit;
int b[] = {1,2,3}; int b[] = {1,2,3};
exit(1, 17, a+b); int a = f(1, 17, b, 36, 55.0, 22.3f);
float *f = g(17.2, b);
} }
/* int f(int x, int y[3]) { */ /* int f(int x, int y[3]) { */

View file

@ -24,6 +24,20 @@ function token_is_type
if b != 0 goto return_1 if b != 0 goto return_1
goto return_0 goto return_0
function functype_return_type
argument ftype
local type
local p
type = ftype + 1
:ftype_rettype_loop
p = types + type
if *1p == 0 goto ftype_rettype_loop_end
type += type_length(type)
goto ftype_rettype_loop
:ftype_rettype_loop_end
return type + 1
; NB: this takes a pointer to struct data, NOT a type ; NB: this takes a pointer to struct data, NOT a type
; Returns 1 if it's a union OR a struct with 1 member (we don't distinguish between these in any way) ; Returns 1 if it's a union OR a struct with 1 member (we don't distinguish between these in any way)
function structure_is_union function structure_is_union
@ -201,6 +215,7 @@ function parse_toplevel_declaration
out = function_stmt_data + function_stmt_data_bytes_used out = function_stmt_data + function_stmt_data_bytes_used
out0 = out out0 = out
ident_list_add(function_statements, name, out) ident_list_add(function_statements, name, out)
ident_list_add(function_types, name, type)
; deal with function parameters ; deal with function parameters
p = type + 1 p = type + 1
@ -1784,6 +1799,7 @@ function parse_type_declarators
:ftype_has_parameters :ftype_has_parameters
:function_type_loop :function_type_loop
if *1p == SYMBOL_DOTDOTDOT goto ftype_varargs
param_base_type = p param_base_type = p
param_prefix = type_get_base_end(param_base_type) param_prefix = type_get_base_end(param_base_type)
param_prefix_end = type_get_prefix_end(param_prefix) param_prefix_end = type_get_prefix_end(param_prefix)
@ -1817,6 +1833,11 @@ function parse_type_declarators
if *1p != SYMBOL_COMMA goto parse_typedecls_bad_type if *1p != SYMBOL_COMMA goto parse_typedecls_bad_type
p += 16 p += 16
goto function_type_loop goto function_type_loop
:ftype_varargs
; ignore varargs
p += 16
if *1p != SYMBOL_RPAREN goto stuff_after_ftype_varargs
goto function_type_loop_end
:function_type_loop_end :function_type_loop_end
if param_names_out == 0 goto ftype_skip_zpno if param_names_out == 0 goto ftype_skip_zpno
*1param_names_out = 0 *1param_names_out = 0
@ -1834,6 +1855,11 @@ function parse_type_declarators
:str_no_param_name :str_no_param_name
string Function parameter has no name. string Function parameter has no name.
byte 0 byte 0
:stuff_after_ftype_varargs
token_error(p, .str_stuff_after_ftype_varargs)
:str_stuff_after_ftype_varargs
string Stuff after ... (varargs) in function type.
byte 0
:parse_type_remove_parentheses :parse_type_remove_parentheses
; interestingly: ; interestingly:
; int (f(int x)) { return x * 2; } ; int (f(int x)) { return x * 2; }
@ -2551,12 +2577,37 @@ function parse_expression
bad_types_to_operator(tokens, *4a, *4b) bad_types_to_operator(tokens, *4a, *4b)
:parse_call :parse_call
*4type = TYPE_INT ; @TODO: proper typing local arg_type
local param_type
; type call
b = types + *4a
if *1b == TYPE_FUNCTION goto type_call_cont
if *1b != TYPE_POINTER goto calling_nonfunction
b += 1 ; handle calling function pointer
if *1b != TYPE_FUNCTION goto calling_nonfunction
:type_call_cont
b -= types
*4type = functype_return_type(b)
param_type = b + 1
:call_args_loop :call_args_loop
if *1p == SYMBOL_RPAREN goto call_args_loop_end if *1p == SYMBOL_RPAREN goto call_args_loop_end
n = token_next_semicolon_comma_rbracket(p) n = token_next_semicolon_comma_rbracket(p)
print_tokens(p, n) arg_type = out + 4
out = parse_expression(p, n, out) out = parse_expression(p, n, out)
b = types + param_type
if *1b == 0 goto arg_is_varargs ; reached the end of arguments (so presumably this function has varargs)
; set argument type to parameter type. this is necessary because:
; float f(float t) { return 2*t; }
; float g(int x) { return f(x); } <- x passed as a float
*4arg_type = param_type
param_type += type_length(param_type)
goto call_arg_type_cont
:arg_is_varargs
type_promote_float_to_double(*4arg_type)
type_decay_array_to_pointer(*4arg_type)
:call_arg_type_cont
p = n p = n
if *1p == SYMBOL_RPAREN goto call_args_loop_end if *1p == SYMBOL_RPAREN goto call_args_loop_end
if *1p != SYMBOL_COMMA goto bad_call if *1p != SYMBOL_COMMA goto bad_call
@ -2565,9 +2616,16 @@ function parse_expression
:call_args_loop_end :call_args_loop_end
p += 16 p += 16
if p != tokens_end goto stuff_after_call if p != tokens_end goto stuff_after_call
*8out = 0 *8out = 0
out += 8 out += 8
return out return out
:calling_nonfunction
token_error(p, .str_calling_nonfunction)
:str_calling_nonfunction
string Calling non-function.
byte 0
:bad_call :bad_call
token_error(p, .str_bad_call) token_error(p, .str_bad_call)
:str_bad_call :str_bad_call
@ -2882,13 +2940,21 @@ function parse_expression
:not_global :not_global
; it must be a function ; it must be a function
c = ident_list_lookup(function_types, a)
if c == 0 goto undeclared_function
*1out = EXPRESSION_FUNCTION *1out = EXPRESSION_FUNCTION
out += 4 out += 4
*4out = TYPE_POINTER_TO_VOID *4out = c
out += 4 out += 4
*8out = a *8out = a
out += 8 out += 8
return out return out
:undeclared_function
; @NONSTANDARD: C89 allows calling functions without declaring them
token_error(in, .str_undeclared_function)
:str_undeclared_function
string Undeclared function.
byte 0
:found_local_variable :found_local_variable
; it's a local variable ; it's a local variable
@ -3011,6 +3077,17 @@ function type_decay_array_to_pointer
type_copy_ids(dest, src) type_copy_ids(dest, src)
return return
; change type to `double` if it's `float`
; in C, float arguments have to be passed as double for varargs
; there is also a rule that char/short/int are passed as ints, but we don't need to worry about it since we're passing everything as >=8 bytes.
function type_promote_float_to_double
argument type
local p
p = types + type
if *1p != TYPE_FLOAT goto return_0
*1p = TYPE_DOUBLE
return
function type_sizeof function type_sizeof
argument type argument type
local p local p