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
61
62
63
64
65
66
|
~ lexer
-- current node in tree is null
-- root node in tree is null
-- amount of lines in tree is incorrect - does this mean we do some kind of weird jumping? >> I think this is fixed
-- does a new line count as a whitespace?
~ parser
~ modules
-- should contain a list of methods and references
-- we want to disable parts of the imports to keep the bloat minimal
-- so bassically compile away the stuff we don't want
-- modules table? list of all available .jr files in a current directory
-- is compiling everything as a single buffer a good idea? or is that just a give me naespace issues cry
~ base
fix the arena implementaiotn
~ The parser's stack request can't fit in the arena. arena_to_stack asserts at base_stack.cpp:223: it wants 10 MiB (requested_size=10485760) from an arena that is only 10 MiB total and already partly used.
Make the arena bigger in compiler.cpp:23 (e.g. MiB(256), or GiBs; untouched mmap pages cost nothing), or ask for less.
parser.cpp:50 still passes arena->current_position as the alignment argument. It only passed the is_pow check by luck and got clamped to 128. Pass arena_align instead.
~ arena_to_stack will crash right after that (base_stack.cpp:201-238).
Lines 227 and 231 use current_position (an offset) as an address. Writing through it segfaults.
Lines 234-235 compute capacity as an address minus an offset. It should be requested_size.
Line 236 sets base_position to an offset.
It never advances arena->current_position, so the next PushStruct overwrites the stack.
Easiest fix: rewrite it on top of arena_alloc, using the sketch from my previous message.
~ The preprocessor reads the wrong module name. That's where opendir(r\n\nenum) failed in your output comes from.
In preproc.cpp:~40, str8_chop_left_ptr(buffer, end_offset) already removed import. The import branch then chops substr.size (6 chars) again, so " std.jr\n\nenum…" becomes "r\n\nenum…".
The name search looks for " " only, so it runs past the newline. It should stop at any whitespace.
consume_token_from_buffer(arena, original_buffer) chops main's buffer in place. After preprocessing, the lexer never sees anything before #import. Pass a copy (dest_buffer is already made and never used).
In scan_buffer_and_handle_modules, buffer = FileLoad(arena, current->value); runs right after the if/else and throws away the str8_append result.
~ tests/base/test.cpp:28 is testing the wrong idea.
ArenaResize(arena, arena->base_position, MiB(1), arena->capacity) tries to resize the whole arena. It fails because nothing is allocated yet (current_position = 0). Rewrite it to test a block: PushStruct a few Foos, resize the last one (in place) and a middle one (copy), then check the contents survived.
Allocator correctness
align() is inverted (base_memory.cpp:5): it returns early for every power of two, so nothing is ever aligned. Change == 0 to != 0.
is_pow(0) returns true. It should be x && (x & (x - 1)) == 0.
arena_resize_block_align (base_arena.cpp) has several problems:
It returns MemArena * and should return void *.
The copy branch never returns new_memory and falls through to NULL.
MemSet(&arena->current_position, ...) overwrites the arena header.
The "last allocation" check uses previous_position, which is unaligned. Use old_mem + old_size == base + current_position.
~ performance
is mmaping faster in reading files?
-- do we pop the buffers fromt mem or not?
|
