1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
|
+ Refactor ast.c and give each operation its own proc, i.e. instead of huge
node_type switches call a corresponding procedure. Should help with debugging,
stack traces etc.
+ Add const and volatile attributes to types
+ Add const to procedures
+ Add external to procedures, essentially disables templates and name mangling,
to be used to interface with C
+ Come up with some better file extension, .cu seems to be used by cuda
- In fact Copper is already taken :(
+ Build new type system (current one build on top of AST fucking sucks)
+ Build the consteval stuff
+ Use built in location tracking with @$ in bison?
+ For now it might be a good idea to just implement generic types to get
started, and add stuff like lambdas and so on later?
OPTIMIZATIONS:
+ Use hashmaps or something similar for lookups for types and calls
+ Output one procedure at a time through the backend, would allow us to free
actualized procs immediately instead of keeping them around in memory
+ Simplify IDs to just char arrays, would save on memory
+ Types can just be types, they maybe don't need to be full on AST nodes
LOWERING:
Structure handling in lowering could probably be as follows:
Imagine
struct something {
i27 a;
other_struct b;
};
main ()
{
// something s = {...};
i27 s_def = alloc $STRUCT_SIZE;
// i27 a = s.a;
i27 member_a = struct_def + 0;
i27 a << struct_def 0;
// other_struct c = s.b;
i27 c_def = alloc $OTHER_STRUCT_SIZE;
i27 member_b = struct_def + 3;
c_def blit member_b, $OTHER_STRUCT_SIZE; // has to be added to qbt,
// might need to be handled
// like store with constant as
// 'return'. Or can c_def be
// considered an output value
// for blit?
/* calls just dump all registers down I guess, as do returns. No
* implicit conversion to pointer and passed as first argument, ufcs
* does that for us. */
}
|
