boot/ is src/ now. The C++ compiler and CMake are gone from this branch (they are on before-bootstrap).
bootstrap/build.sh builds the compiler from bootstrap/<system>.ll -- the compiler's own LLVM IR -- then builds
it with itself and checks the two come out identical. bootstrap/build.sh seed writes the seed anew.
The driver is complete: the runtime is compiled once and kept in ~/.cache/finch; a program's own C files,
.o and .a files and frameworks link along; a linker failure is read and answered with link "lib" or the
library that is missing. The old tests/fail diagnostics are all in again (a missing import, a str compared with
null, 'write just the library's name', a map key that is a float, non-ASCII names, ...).
tests/run.sh is the one test runner: programs, refusals, -g builds and the shared library. 106 passed.
Targets, the target's C compiler (FINCH_CC) for linking, C import for the target (--target to libclang, its
headers), the AAPCS64 and Microsoft x64 ways of passing structs by value (and the System V one),
dllexport for libraries on Windows, .exe names, and running the program through wine or qemu-aarch64.
std/ gets time_windows.fch and proc_windows.fch: a module has a version per system (name_windows.fch,
name_macos.fch, name_linux.fch) that wins over name.fch. stdin, stdout and stderr work where C makes them macros.
Run for real: all 27 programs under tests/run and tests/new pass on ARM64 Linux under qemu and on Windows
under Wine (MinGW), structs by value and callbacks included. macOS is only checked as far as clang accepts the
IR; there is no emulator for it here.
DWARF as LLVM metadata: a compile unit, a subprogram per function, a location on every call, a variable
for each local and parameter with types a debugger can open (structs with their fields, arrays and text
as ptr/len/cap, maps, enums with their variants, pointers, fixed arrays). Methods show self.
Use it with -O0, as the docs say: the optimizer inlines the program into main, which has no debug info.
boot.sh builds ten programs with -g -O0 and checks they still behave. 51 passed.
export fn name(...) is a C function of a library; build --shared makes lib<name>.so with no main and a
finch_module_on_fatal(handler) the host can take runtime errors over with. Ported from the embed work (the
runtime, its test and the docs are that work's; the compiler side is new).
The driver reads FINCH_PATH (more module folders), FINCH_RT (the runtime's file) and FINCH_CC (the clang).
tests/boot.sh also runs tests/shared.sh with the compiler under test. 41 passed.
import json import path import time import proc -- found in std/ after the program's own files,
so a program's own json.fch always wins.
path: join, base, dir, ext, stem, parts, clean, is_abs.
time: now, millis, sleep, date, local, format, days_in_month (clock_gettime, strftime through C import).
proc: run, output (an error with the command's output when it fails), env, pid, cwd.
json: parse (errors with line and column), stringify, pretty, new_*, get/at/set/push/has/names and
str_or/int_or/float_or/bool_or; \u escapes with surrogate pairs.
The C importer now leaves a plain (non-const) char * parameter a ptr[char], a buffer C writes into;
only const char * is text going in. str(ptr) and str([]u8) are in the self-hosted compiler.
boot.sh: 40 passed, 0 failed.
fn max[T](T a, T b) -> T struct Stack[T] { []T items } fn Stack[T].push(T x)
The type is worked out from the arguments (max(3, 4), first([1, 2])) and written out
where it cannot be: max[float](1, 2.5), Stack[int](), Pair[str, int](...). Each use makes
its own copy, so a generic costs nothing at run time. Several parameters
(Pair[A, B]), nesting (Stack[Pair[int, str]]), methods, and printing (Stack[int](items: ...))
work. An error inside a generic says which version it was building.
boot.sh: 37 passed, 0 failed (stage 2 and 3 identical).
enum Color { Red, Green, Blue } enum Status { Ok = 200, NotFound = 404 }
Color.Red, Color(n) (checked), int(c), == and != only, printed by name, usable as
map keys and struct fields, with methods (fn Color.next()).
match value { Red, Green -> ... Blue -> ... else -> ... } as a statement, or as an
expression that gives a value. Works on enums, whole numbers, chars, text and bools;
a match on an enum has to handle every variant or end with else, and says which are
missing. A bare variant name (Red) is enough inside a match on an enum.
Only the self-hosted compiler has them; tests/new holds their tests, tests/run stays what
the C++ compiler also runs. boot.sh: 31 passed.
C structs travel as C passes them: up to 16 bytes in INTEGER/SSE registers per eightbyte,
bigger ones byval and through sret, out-of-registers structs on the stack. addr(fn) makes a
C-callable thunk that rebuilds the arguments. With those, the self-hosted compiler can
compile boot/cimport.fch, which calls libclang with CXCursor and CXString by value and gives
it callbacks -- so it compiles itself.
boot.sh: stage 2 and stage 3 are identical (57904 lines of IR); 22 passed, 0 failed,
0 skipped. Only x86-64 System V so far; Windows and ARM64 conventions are still in C++.
The self-hosted compiler reads headers with boot/cimport.fch and compiles calls to C
functions (variadics, char* as str, C's default promotions), enum and #define
constants, extern variables like stdout, C structs laid out as C laid them (padding
and fixed arrays, by pointer), .ptr of a str or array, ptr(x) and int(ptr). link
"lib" asks pkg-config, link "x.c" compiles the file along.
cimport and c_pointers pass. Not yet: structs by value (the calling convention) and
addr(fn) for C callbacks, which is also what stops it building itself.
boot.sh now says so when stage 2 can't be made and runs the tests on stage 1.
map[K]V, [:] and ["a": 1] literals, m[k] reads and writes (m[k] += 1 starts from
zero), has/get/remove/clear/keys/values, .len, copy and printing, and a loop over
a map that stops with an error when it adds keys. for i, x in list gives the
index too.
boot.sh: 17 passed, 0 failed, 5 skipped (all five import C).
A function that can fail returns { ok, value, message }. return error(...) makes
the failure; try passes it on after running the defers; 'or <value>' and
'or { ... err ... }' handle it. int(text), float(text), read_file, write_file
and delete_file report failure to an enclosing or/try instead of stopping the
program. fn main() -> ! prints the error and exits with 1.
boot.sh: 15 passed, 0 failed, 7 skipped.
defer runs when its scope ends and before return, break and continue leave it,
newest first, and sees only the variables declared before it. Not allowed
inside a defer: defer, return, break, continue.
boot.sh: 14 passed, 0 failed, 8 skipped.
Methods are functions whose first parameter is the receiver's address; a method
that changes self can't be called on a temporary or a read-only value, and
calling one on a variable keeps a parameter from being borrowed. An argument
that lives inside the receiver is copied (c.merge(c)). str(x) and print of
arrays and structs go through one formatter, as in the C++ compiler.
boot.sh: 12 passed, 0 failed, 10 skipped.
The self-hosted compiler now has the numeric types the C++ one has: int and
float stay the i64 and f64 defaults, the others lower to their LLVM sizes;
literals adapt to the other operand if they fit, smaller types widen into
bigger ones, signed and unsigned mixing is refused, division and shifts and
comparisons pick their signedness, and u8(x), f32(x), bool(x) convert.
boot.sh: 11 passed, 0 failed, 10 skipped (was 9 / 12).
.fn is already used by other tools (e.g. Fennel). Modules are now looked up as name.fch;
a leftover name.fn gets a clear message telling you to rename it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
