l0ng-ai e80fc24ce0 test(protocol): round-trip every wire message, and keep it that way ...

`client_roundtrip` and `daemon_roundtrip` are hand-written lists, so a
variant added later goes untested unless somebody remembers it — and nothing
said otherwise. Eleven had drifted out of the two lists; five of those
(`Handoff`, `QueryProcs`, `Image`, `DeleteImage`, `Procs`) were not encoded
or decoded by *any* test in the module. `Image` is the one carrying arbitrary
binary through a length-prefixed frame, which is exactly where a framing bug
would hide.

All eleven are added and all of them round-trip; nothing was broken. `Image`
gets 70,000 bytes cycling through every value, so it crosses the read
buffer, plus an empty one; `DeleteImage` gets a payload containing the APC
bytes a naive framer might resynchronise on; `Procs` gets both a populated
`PaneProcs` and an empty one; `AuthResponse` and `SshStatus` get each of
their own shapes.

The guard reads this file's own text: the variants `pub enum ClientMsg` and
`pub enum DaemonMsg` declare, against the ones the matching test names. One
list instead of two, for the usual reason. It asserts it parsed more than
twenty variants first, so a change in how the file is written fails loudly
rather than passing for the wrong reason.

Note what this does *not* need to catch: a brand-new variant does not
compile until `encode` and `read` handle it, so the compiler has that half.
What it catches is the half the compiler cannot — a variant that is wired up
correctly and simply never tested. Verified by dropping `QueryProcs` back out
of the list, which it names.

2026-08-23 03:31:12 +08:00

..

tty7-cli

docs(cli): server restart keeps sessions, and --hard is the one that does not

2026-08-23 00:58:36 +08:00

tty7-core

test(protocol): round-trip every wire message, and keep it that way

2026-08-23 03:31:12 +08:00

tty7-server

fix(history): stop per-pane history from silently disabling history search

2026-08-23 02:46:54 +08:00