Spec M3 — globals, global arrays, strings, #include, #define, arena, data sections

Prerequisite: M2 closed and reviewed. Scope: everything a real program (the M4 tokenizer and later mc.mc) needs in terms of static data and modularization.

Parser #

• Top level: type name = CONST; (N_GLOBAL, a=init folded to N_INT), type name;, type name[CONST]; (val=nelem). A non-constant initializer is an error.
• #include "path": relative to the directory of the including file; once-only per resolved path (linear list, insertion order). Implement as a stack of files in the lexer (push on include, pop on EOF). Errors show file:line of the correct file. Max depth 16.
• #define NAME expr: expr parsed and folded at definition time; linear table {name, val}. When the parser reads an identifier in an expression it checks the table before anything else and emits N_INT. Redefining = error. #define must come before its use (source order).
• String literals in expressions (N_STR: bytes + len in the arena).

Codegen #

• Sections created on demand, always in this order when present: __TEXT,__text (align 2), __TEXT,__cstring (S_CSTRING_LITERALS, align 0), __DATA,__data (align 3), __DATA,__bss (S_ZEROFILL, align 4, zsize).
• Globals: symbol _name not external; scalars in __data (initialized, type width) or in __bss (no initializer); arrays always in __bss, size nelem*width rounded to 16, aligned to 16. Access: adrp xd, _name@PAGE (R_PAGE21, pcrel=1, len=2) + add xd, xd, _name@PAGEOFF (R_PAGEOFF12, pcrel=0, len=2) then ldr/str with the type's width. A global array as a value / &global = the address after the add.
• Strings: bytes + NUL in __cstring; local symbol l_str<N> (N sequential in order of first occurrence); dedup by content via linear search. Value = address via adrp/add, type uptr.
• Name resolution in an expression: define (parser) → local → global → function/extern (for calls).
• --dump-syms: sections (seg,sect,flags,size) and symbols in symtab final order, one per line.

Library #

• lib/sys.mc: extern i64 open(uptr path, i64 flags, i64 mode); extern i64 read(i64 fd, uptr buf, i64 n); extern i64 write(i64 fd, uptr buf, i64 n); extern i64 close(i64 fd); extern void exit(i64 code); and #define O_RDONLY 0, #define O_WRONLY 1, #define O_CREAT 0x200, #define O_TRUNC 0x400 (macOS values). Also puts(uptr s) (strlen + write) and putnum(i64), written in .mc.

Tests #

• 020-globals.mc: global counter incremented inside a function, global array u64[8] → 42.
• 021-strings.mc: #include "../lib/sys.mc"; puts("hello\n") → stdout hello\n, exit 0.
• 022-include.mc: includes tests/lib/util.mc (create it) twice (once-only) and uses one of its functions.
• 023-define.mc: #define with a constant expression used both in an array size and in an expression → 42.
• 024-arena.mc: u8 heap[4096]; i64 hp = 0; uptr alloc(i64 n); allocates two blocks, st64/ld64, checks alignment to 8 → 42.
• 025-linecount.mc: opens tests/025-linecount.mc (path relative to the repo root, where test.sh runs), reads in a loop of 4096-byte chunks into a global buffer, counts \n, prints with putnum → stdout = the number of lines in the file itself.

Acceptance #

make stage0 && make stage0-san && make budget && make test green (all M1–M3 tests); otool -r and nm -m of the 021-strings .o showing l_str0 in __cstring and PAGE21/ PAGEOFF12 relocs; --dump-syms deterministic. Report lines per file and the total.

Edit this page