7612ce44

docs: deinit comment takes both review nits; roadmap strikes the reorder

a73x 2026-08-13 09:43

Commit message

docs: deinit comment takes both review nits; roadmap strikes the reorder

Two Minor findings from the quality review of c492eb2, both wording:
the successor concludes nothing is LISTENING (claim()'s probe cannot
refuse while our fd is open), not that our process exited; and the
stat→unlink gap is named as the floor for deleting by name — Linux has
no inode-predicated unlink. The roadmap entry is struck with the sha
and carries the review's sharpest finding: the old stat-before-close
guard could never fire in the very race it guarded.

Closes: 2cb6f950

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

docs/roadmap.md

Old		New
@@ -190,10 +190,12 @@ next one that arrives outranks all of it.
190	- ~~The transport recipe as a `union(enum)`~~ — paid by M15 (Task 6).	190	- ~~The transport recipe as a `union(enum)`~~ — paid by M15 (Task 6).
191	`Transport.open` takes `union(enum) { sock, via, quic, hand }`, and the	191	`Transport.open` takes `union(enum) { sock, via, quic, hand }`, and the
192	dispatch decision it makes got its first pin in the same task.	192	dispatch decision it makes got its first pin in the same task.
193	- The stat-after-close reorder in `deinit` — created by M15 (Task	193	- ~~The stat-after-close reorder in `deinit`~~ — paid (M16-b,
194	10). Narrows the window in which a departing daemon's cleanup could	194	c492eb2). Review found the old order was worse than vestigial: a
195	unlink a successor's socket. The analysis is already written, in the	195	successor cannot pass `claim()` while our listener fd is open, so the
196	`sockpath` comment; only the reorder is owed.	196	stat-before-close guard could never fire in the very race it guarded.
		197	The residual stat→unlink gap is the floor — Linux has no
		198	inode-predicated unlink.
197	- `egress_cap` ↔ transport-params, one owner — also M15. `256*1024`	199	- `egress_cap` ↔ transport-params, one owner — also M15. `256*1024`
198	is spelled three times and the tie between the copies is prose only.	200	is spelled three times and the tie between the copies is prose only.
199	- Fold `Transport`'s `alloc`/`qout` into the `.quic` `Link` payload —	201	- Fold `Transport`'s `alloc`/`qout` into the `.quic` `Link` payload —

src/server.zig

Old		New
@@ -336,12 +336,15 @@ pub const Server = struct {
336	// The stat comes AFTER the close. Statting first was vestigial:	336	// The stat comes AFTER the close. Statting first was vestigial:
337	// 8a82225 fstat'd the live descriptor and so needed the fd open,	337	// 8a82225 fstat'd the live descriptor and so needed the fd open,
338	// 6090604 switched to statting the path and killed that constraint.	338	// 6090604 switched to statting the path and killed that constraint.
339	// A successor only claims the path after concluding we are gone, so	339	// A successor only claims the path after concluding nothing is
340	// asking on this side of our close is the correct side of that event,	340	// listening here — claim()'s probe cannot refuse while our fd is
341	// and it shrinks the window in which one can claim the path between	341	// open — so asking on this side of our close is the correct side of
342	// our answer and our unlink down to the stat→unlink gap. The close	342	// that event, and it shrinks the window in which one can claim the
343	// itself stays unconditional: the fd is ours whoever owns the path,	343	// path between our answer and our unlink down to the stat→unlink
344	// and only the unlink touches the shared namespace.	344	// gap. That gap is the floor for deleting by name: Linux has no
		345	// inode-predicated unlink. The close itself stays unconditional:
		346	// the fd is ours whoever owns the path, and only the unlink touches
		347	// the shared namespace.
345	self.listener.deinit();	348	self.listener.deinit();
346	if (self.path_id.stillAt(self.sock_path)) {	349	if (self.path_id.stillAt(self.sock_path)) {
347	std.fs.cwd().deleteFile(self.sock_path) catch {};	350	std.fs.cwd().deleteFile(self.sock_path) catch {};