Phase 0 last β-shape site. Two concerns in one commit because the refactor surfaced the rename: - selfhost/cmd/wcc/cgen.ww cgout buffer (cgoutbuf/cap/len + arena + cgout_grow + CGOUT_INIT_CAP) → memio.state + io.stream behind a one-shot lazy-init guard. cgout_enable drops its *arena param; memio.reset in cgout_flush keeps the buffer sticky across fns so the arena's amortisation survives — re-init per fn would abandon the buffer and re-grow from 0 via the 8→…→65536 ladder for every function (no io.close path → no os.free). - lib/memio/memio.ww private fn grow → dynamicgrow. Symmetric with dynamicwrite / dynamicclose; required because cstage bundles all imported modules into a flat TU and resolves private fns by unqualified name, so the new `import memio;` in wcc's bundle collided with selfhost/cmd/wcc/mem.ww's arena `grow`. Module-aware private-fn scoping in cstage is task #9. @test fn dynamicgrow in memiotest.ww (same package as memio.ww) renamed to dynamicgrowcases to free the name; new suffix mirrors the file's existing fixedwritecases / borrowedreadcases convention. Lazy-init guard cgoutinit. memio.dynamic runs once on first cgout_enable; subsequent enables just set cgoutmode. Mirrors lib/log/log.ww:124 ensureinit. Without it, ~14 mmap syscalls per fn and ~100 MiB+ cumulative leak on a typical bootstrap. io.write bare discard in emitbytes mirrors lib/log/log.ww:169 — memio.dynamicwrite never returns io.closed (memio.ww:166). Verified 132/132 incl. 995_self_rebuild byte-identity.
217 lines
6.1 KiB
Plaintext
217 lines
6.1 KiB
Plaintext
// memio — in-memory io stream.
|
|
//
|
|
// Hare's memio:: surface, drop underscores. Two flavours behind a
|
|
// single [[io.stream]]:
|
|
//
|
|
// fixed caller owns the buffer, writes stop when full.
|
|
// dynamic memio owns the buffer, writes grow it; close frees.
|
|
//
|
|
// Call shape divergence from Hare: the caller supplies both the
|
|
// memio `state` and the `io.stream` slot, by pointer. ww cgen does
|
|
// not yet implement &x.field or 32B-struct return-by-value, so the
|
|
// Hare `let s = memio::fixed(buf)` shape isn't reachable; collapse
|
|
// to a single returned struct when those land (lib/CLAUDE.md
|
|
// "graduate in one go").
|
|
//
|
|
// let mem: memio.state;
|
|
// let s: io.stream;
|
|
// memio.fixed(&mem, &s, buf);
|
|
// io.write(&s, bytes);
|
|
//
|
|
// Subset of Hare's surface: io.stream's variants are {eof, closed},
|
|
// so memio drops Hare's NONBLOCK flag (would need an `again` variant
|
|
// in lib/io). string()'s utf8-validating constructor is omitted per
|
|
// CLAUDE.md rule 9 carve-out. Hare's seek / copy callbacks are
|
|
// likewise absent: lib/io's stream vtable has only read/write/close
|
|
// slots, so memio can't wire a seeker or copier even if we wanted
|
|
// to. All three come back when their dependencies do.
|
|
|
|
package memio;
|
|
|
|
import io;
|
|
import os;
|
|
import rt;
|
|
|
|
// state — memio's per-stream bookkeeping. The caller owns the slot
|
|
// and passes its address into a constructor. `ptr/len/cap` are the
|
|
// slice fields kept flat to dodge a chained-dot write through the
|
|
// state pointer (cgen doesn't store into `m.buf.ptr` reliably).
|
|
export type state = struct {
|
|
ptr: *u8,
|
|
len: i32,
|
|
cap: i32,
|
|
pos: i32,
|
|
};
|
|
|
|
// fixed — wire `s` over a caller-supplied buffer. Writes never grow;
|
|
// they return 0 once `pos` reaches the end of the buffer.
|
|
export fn fixed(m: *state, s: *io.stream, buf: []u8) void = {
|
|
m.ptr = buf.ptr;
|
|
m.len = buf.len;
|
|
m.cap = buf.len;
|
|
m.pos = 0;
|
|
s.ctx = m: *void;
|
|
s.read = readfn;
|
|
s.write = fixedwrite;
|
|
s.close = closenoop;
|
|
};
|
|
|
|
// dynamic — wire `s` with no initial buffer. Writes grow the backing
|
|
// allocation; [[io.close]] frees it.
|
|
export fn dynamic(m: *state, s: *io.stream) void = {
|
|
m.ptr = nil;
|
|
m.len = 0;
|
|
m.cap = 0;
|
|
m.pos = 0;
|
|
s.ctx = m: *void;
|
|
s.read = readfn;
|
|
s.write = dynamicwrite;
|
|
s.close = dynamicclose;
|
|
};
|
|
|
|
// dynamicfrom — like [[dynamic]] but seeded with an existing slice.
|
|
// Ownership of the slice transfers to the stream; [[io.close]] frees
|
|
// it. The slice must come from the runtime allocator: close calls
|
|
// [[os.free]] with `m.cap` bytes, which is taken from `buf.cap` (the
|
|
// slice's allocated capacity), not its logical length. Passing a
|
|
// half-filled append slice (len < cap) and using only `buf.len` here
|
|
// would under-free on close.
|
|
export fn dynamicfrom(m: *state, s: *io.stream, buf: []u8) void = {
|
|
m.ptr = buf.ptr;
|
|
m.len = buf.len;
|
|
m.cap = buf.cap;
|
|
m.pos = 0;
|
|
s.ctx = m: *void;
|
|
s.read = readfn;
|
|
s.write = dynamicwrite;
|
|
s.close = dynamicclose;
|
|
};
|
|
|
|
// buffer — borrowed view of bytes written so far (buf[..pos]).
|
|
// Seek to the end before calling if the full buffer is wanted.
|
|
export fn buffer(m: *state) []u8 = {
|
|
let r: []u8;
|
|
r.ptr = m.ptr;
|
|
r.len = m.pos;
|
|
return r;
|
|
};
|
|
|
|
// string — bytes written so far, as a str view. Hare returns
|
|
// (str | utf8::invalid); ww doesn't ship utf8 validation yet, so
|
|
// this returns the unchecked view.
|
|
export fn string(m: *state) str = {
|
|
let r: str;
|
|
r.ptr = m.ptr;
|
|
r.len = m.pos;
|
|
return r;
|
|
};
|
|
|
|
// reset — rewind the cursor and truncate the logical content to 0.
|
|
// Backing storage is preserved; subsequent writes (dynamic) re-fill
|
|
// from the start without reallocation.
|
|
export fn reset(m: *state) void = {
|
|
m.pos = 0;
|
|
m.len = 0;
|
|
};
|
|
|
|
// borrowedread — return an `amt`-byte view starting at `pos` without
|
|
// copying, advancing the cursor. eof if fewer bytes are available.
|
|
export fn borrowedread(m: *state, amt: i32) ([]u8 | io.eof) = {
|
|
if (m.len - m.pos < amt) {
|
|
let e: io.eof;
|
|
return e;
|
|
};
|
|
let r: []u8;
|
|
r.ptr = m.ptr + (m.pos: u64);
|
|
r.len = amt;
|
|
m.pos += amt;
|
|
return r;
|
|
};
|
|
|
|
// ---- vtable callbacks ------------------------------------------------
|
|
|
|
fn readfn(s: *io.stream, buf: []u8) (i32 | io.eof | io.closed) = {
|
|
let m: *state = s.ctx: *state;
|
|
if (m.pos >= m.len) {
|
|
let e: io.eof;
|
|
return e;
|
|
};
|
|
let avail: i32 = m.len - m.pos;
|
|
let n: i32 = buf.len;
|
|
if (avail < n) { n = avail; };
|
|
let i: i32 = 0;
|
|
for (i < n) {
|
|
buf[i] = m.ptr[m.pos + i];
|
|
i += 1;
|
|
};
|
|
m.pos += n;
|
|
return n;
|
|
};
|
|
|
|
fn fixedwrite(s: *io.stream, buf: []u8) (i32 | io.closed) = {
|
|
let m: *state = s.ctx: *state;
|
|
if (m.pos >= m.len) { return 0; };
|
|
let space: i32 = m.len - m.pos;
|
|
let n: i32 = buf.len;
|
|
if (space < n) { n = space; };
|
|
let i: i32 = 0;
|
|
for (i < n) {
|
|
m.ptr[m.pos + i] = buf[i];
|
|
i += 1;
|
|
};
|
|
m.pos += n;
|
|
return n;
|
|
};
|
|
|
|
fn dynamicwrite(s: *io.stream, buf: []u8) (i32 | io.closed) = {
|
|
let m: *state = s.ctx: *state;
|
|
let need: i32 = m.pos + buf.len;
|
|
if (need > m.cap) { dynamicgrow(m, need); };
|
|
let i: i32 = 0;
|
|
for (i < buf.len) {
|
|
m.ptr[m.pos + i] = buf[i];
|
|
i += 1;
|
|
};
|
|
m.pos += buf.len;
|
|
if (m.pos > m.len) { m.len = m.pos; };
|
|
return buf.len;
|
|
};
|
|
|
|
fn dynamicclose(s: *io.stream) (void | io.closed) = {
|
|
let m: *state = s.ctx: *state;
|
|
if (m.cap > 0) { os.free(m.ptr: *void, m.cap: u64); };
|
|
m.ptr = nil;
|
|
m.len = 0;
|
|
m.cap = 0;
|
|
m.pos = 0;
|
|
return;
|
|
};
|
|
|
|
fn closenoop(s: *io.stream) (void | io.closed) = {
|
|
return;
|
|
};
|
|
|
|
// Double-and-copy growth. Initial bump from 0 lands at 8 to amortise
|
|
// small write bursts without a tail of reallocs.
|
|
//
|
|
// `dynamicgrow`, not Hare's bare `grow`: cstage bundles all imported
|
|
// modules into a flat TU and resolves private fns by unqualified
|
|
// name, so two `fn grow` decls (here + selfhost/cmd/wcc/mem.ww's
|
|
// arena `grow`) collide. Module-prefixed name keeps the symmetry
|
|
// with `dynamicwrite`/`dynamicclose` until task #9 (module-aware
|
|
// private-fn scoping in cstage) lands.
|
|
fn dynamicgrow(m: *state, need: i32) void = {
|
|
let newcap: i32 = m.cap;
|
|
if (newcap < 8) { newcap = 8; };
|
|
for (newcap < need) { newcap *= 2; };
|
|
let nbuf: *u8 = rt.malloc(newcap: u64): *u8;
|
|
let i: i32 = 0;
|
|
for (i < m.len) {
|
|
nbuf[i] = m.ptr[i];
|
|
i += 1;
|
|
};
|
|
if (m.cap > 0) { os.free(m.ptr: *void, m.cap: u64); };
|
|
m.ptr = nbuf;
|
|
m.cap = newcap;
|
|
};
|