fix: remove the em-dashes this release added (release review)

105 new em-dashes sat in comments, test names and one test fixture from the
foreign tmux adoption work. Replaced by hand with commas, colons or
parentheses, one reading at a time. The fixture now quotes the real refusal
text, which says "It may still be running, try again."

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Codeman maintainer committed 2026-11-06 08:20:39 +01:00
1 parent 9d2f9663cb
commit 9442ea8cf4
27 files changed
+99 -99

No files matched your search

+27 -27
View File
@@ -1,5 +1,5 @@
/**
* @fileoverview Pure core for FOREIGN tmux sessions — the ones a human started
* @fileoverview Pure core for FOREIGN tmux sessions, the ones a human started
* by hand, which Codeman neither created nor owns.
*
* Everything here is a string in / structure out, so all three locations (local,
@@ -32,13 +32,13 @@
* ⚠️ The script MUST NOT contain a single quote. It is wrapped in single quotes
* to cross `ssh <host> '<script>'` and `docker exec <c> sh -lc '<script>'`. That
* is the single-quote dual of the "no double quotes in the docker launch chain"
* rule in `tmux-manager.ts` — same class of bug, opposite quote.
* rule in `tmux-manager.ts`: same class of bug, opposite quote.
*
* ## Why the mode comes from the process tree
*
* `#{pane_current_command}` is `node` for BOTH claude and codex, so it cannot
* distinguish them. The classifier walks the pane's descendants over the `ps`
* snapshot with `collectDescendants` (bounded: one snapshot, no per-node spawn —
* snapshot with `collectDescendants` (bounded: one snapshot, no per-node spawn,
* see proc-tree.ts for the incident that rule exists for) and matches argv. What
* it cannot identify is `shell`, which is also the honest answer for the case
* this feature was built for: a hand-opened shell.
@@ -53,7 +53,7 @@
*
* ⚠️ The group session must be created ATTACHED (no `-d`). tmux's
* `server_check_unattached()` runs every server loop, so a detached session with
* `destroy-unattached on` is torn down almost immediately — the option has to
* `destroy-unattached on` is torn down almost immediately: the option has to
* land on a session that already has our client on it.
*
* ⚠️ **Grouping does NOT give us an independent window SIZE, and believing it did
@@ -66,12 +66,12 @@
*
* A window is one object with one size; a group shares the object. So attaching
* from Codeman resizes a session someone else is actively viewing, exactly like
* any second tmux client does — that is tmux's normal behavior, and it reverses
* any second tmux client does: that is tmux's normal behavior, and it reverses
* when we detach.
*
* `window-size largest` is deliberately NOT set here. It is a WINDOW option on a
* SHARED window, so it survives our detach (measured: still `largest` after our
* view is gone) — it would permanently rewrite the owner's configuration to buy
* view is gone): it would permanently rewrite the owner's configuration to buy
* a cropped view for us. Restoring it would need the previous value captured with
* `$(...)`, and this string carries no command substitution. It reaches `bash -c`
* verbatim now (`bashCArg()`), but it crossed `bash -c "..."` for a long time, where
@@ -102,7 +102,7 @@ import type { SessionAdopt, SessionMode } from './types/session.js';
/**
* Field separator. tmux's `-F` does NOT expand a backslash-t escape: it emits the
* LITERAL two characters (verified on next-3.7 — the same finding that shaped
* LITERAL two characters (verified on next-3.7, the same finding that shaped
* `parseRemoteSessionList` in remote-hosts.ts). The parser accepts a real TAB too,
* in case some build does expand it.
*/
@@ -145,12 +145,12 @@ export function buildForeignProbeScript(): string {
// `ps -eo` covers procps and macOS; `ps ax -o` is the BSD-ish fallback. If both
// fail we still emit the pane rows, and classification degrades to the pane
// command — which still names a plain shell correctly.
// command, which still names a plain shell correctly.
// ⚠️ No `set -f` here: the socket sweep IS a glob, and disabling pathname
// expansion turns the loop into a single literal non-match. ⚠️ The marker line
// must NOT try to carry the field separator at all. Measured: `sh`'s builtin
// `echo` EXPANDS a backslash-t to a real TAB, while tmux's own `-F` emits the
// two literal characters — so the same escape means two different things two
// two literal characters, so the same escape means two different things two
// lines apart. The socket line is space-separated instead (marker first, rest
// of line is the path), which no shell rewrites.
//
@@ -234,7 +234,7 @@ export function parseForeignProbeOutput(stdout: string): ForeignProbeOutput {
if (inProcs) {
if (procCount >= FOREIGN_MAX_PROCS) continue;
// `<pid> <ppid> <argv...>` — leading spaces are how `ps -o pid=` pads.
// `<pid> <ppid> <argv...>`: leading spaces are how `ps -o pid=` pads.
// ⚠️ `s`, so `.*$` cannot fail: without it a bare CR in an argv stopped the `.`, and
// `\s+` handed its spaces back one at a time, quadratic in a row this long.
const m = /^\s*(\d+)\s+(\d+)\s+(.*)$/s.exec(line);
@@ -305,7 +305,7 @@ export function parseForeignProbeOutput(stdout: string): ForeignProbeOutput {
*
* Matched against the WHOLE argv of a pane descendant, so `node .../bin/claude`
* hits the claude rule via the path. The binary names come from each entry's
* `discovery.binaries` — the one place that already knows a mode name is NOT
* `discovery.binaries`, the one place that already knows a mode name is NOT
* always a binary name (`antigravity` runs `agy`). The hand-written table this
* replaces keyed on CLI ids, which CLAUDE.md forbids outside `stock.ts`, and it
* had already fallen behind: `omp` was missing, so a hand-started omp pane read
@@ -339,7 +339,7 @@ export function foreignCliSignatures(): ReadonlyArray<{ mode: SessionMode; test:
// on purpose: `SessionMode` is the CLOSED union of the stock ids, so a custom
// entry has no SessionMode to be. Letting one through would put a value the
// union does not contain on `SessionState.mode`. A custom CLI's pane falls to
// the `shell` answer below, which is the honest one — Codeman has no name for
// the `shell` answer below, which is the honest one: Codeman has no name for
// it, and an adopted wrapper runs `attach` either way.
if (!entry.stock) continue;
for (const binary of entry.discovery.binaries) {
@@ -413,7 +413,7 @@ export function classifyForeignPaneMode(
if (direct) return { mode: direct, command: truncate(own) };
// ⚠️ Report truncation. A walk that stopped at a cap returns "nothing matched",
// which is byte-identical to the answer for a real shell — without this line a
// which is byte-identical to the answer for a real shell: without this line a
// deep or wide tree silently downgrades an agent pane to `shell` and there is
// nothing anywhere to tell the two apart.
const descendants = collectDescendants(pane.panePid, probe.byParent, {
@@ -503,7 +503,7 @@ function truncate(s: string): string {
}
// ===========================================================================
// Exclusion — never adopt what Codeman already owns
// Exclusion: never adopt what Codeman already owns
// ===========================================================================
/** tmux servers Codeman runs itself. Sessions on these are never "foreign". */
@@ -518,7 +518,7 @@ const CODEMAN_SESSION_PREFIXES = ['codeman-', 'claudeman-'];
* Excluding them is not cosmetic. Locally it stops Codeman from adopting its own
* sessions (which would attach a second PTY to a live pane). On a remote host or
* inside a container it stops the SAME session from being reachable through two
* different code paths — `remote-hosts.listRemoteCodemanSessions` already owns
* different code paths: `remote-hosts.listRemoteCodemanSessions` already owns
* the `tmux -L codeman` case there, and two paths to one session is how the two
* end up disagreeing about ownership.
*
@@ -539,7 +539,7 @@ export function isCodemanOwnedPane(
}
// ===========================================================================
// Adoptability — what may cross the launch chain at all
// Adoptability: what may cross the launch chain at all
// ===========================================================================
/**
@@ -568,8 +568,8 @@ export function isCodemanOwnedPane(
* dropped during discovery, so it has no id, and the adopt endpoint re-resolves
* through that same discovery and can only 404.
*
* Generous on purpose about what real names contain — Unicode letters, digits,
* marks, spaces, and the punctuation people actually use — because a refusal
* Generous on purpose about what real names contain (Unicode letters, digits,
* marks, spaces, and the punctuation people actually use), because a refusal
* here is a session the user cannot open at all.
*/
const ADOPTABLE_TEXT = /^[\p{L}\p{N}\p{M} _.+=@#%,~^!/-]+$/u;
@@ -732,13 +732,13 @@ export interface ForeignAttachTarget {
* target gone tmux takes it as part of the name and creates a group literally
* called `=work`. It has to be a separate `has-session`.
* - `new-session -t` creates a session in the target's GROUP: same windows, our
* own session options and current window (NOT our own size — see @fileoverview
* own session options and current window (NOT our own size, see @fileoverview
* for the measurement that disproved that).
* - `status off` is why grouping earns its keep: it lands on OUR session, so the
* owner's status bar is untouched.
* - No `-d`: the session must be attached before `destroy-unattached` lands on it.
* - The escaped semicolons reach tmux as plain command separators after one shell
* parse — the same convention the docker launch chain uses.
* parse, the same convention the docker launch chain uses.
* - ⚠️ There is NO degraded fallback, on purpose. A read-only `attach-session -r`
* used to hang off `||` "for a tmux too old to group", but `||` fires on ANY
* `new-session` failure. The one that actually happens is a leftover view with
@@ -796,7 +796,7 @@ export function buildForeignAttachCommand(target: ForeignAttachTarget): string {
}
export interface ForeignDockerAttachOptions extends ForeignAttachTarget {
/** `buildDockerBaseArgs(...)` joined — e.g. `docker` or `docker --context foo`. */
/** `buildDockerBaseArgs(...)` joined, e.g. `docker` or `docker --context foo`. */
dockerBase: string;
containerName: string;
}
@@ -805,7 +805,7 @@ export interface ForeignDockerAttachOptions extends ForeignAttachTarget {
* Docker adoption: `docker exec -it` into the container and attach there.
*
* Mirrors the ADOPTED-container branch of `buildDockerLaunchCommand`: look, then
* exec. Never `create`, never `start` — the container belongs to the user, and a
* exec. Never `create`, never `start`: the container belongs to the user, and a
* missing or stopped one fails closed with a message instead of being mutated.
*/
export function buildForeignDockerAttachCommand(opts: ForeignDockerAttachOptions): string {
@@ -833,7 +833,7 @@ export function buildForeignDockerAttachCommand(opts: ForeignDockerAttachOptions
* ⚠️ Needed for docker and NOT for the other two, which is a docker fact rather
* than a design choice. Killing the wrapper pane kills a LOCAL tmux client
* directly, and over ssh it sends SIGHUP to the remote client (the same
* propagation the non-owned remote attach has relied on since COD-105) — either
* propagation the non-owned remote attach has relied on since COD-105), either
* way the client detaches and `destroy-unattached on` collects the view.
* A `docker exec` process does NOT die when its client goes away: measured, the
* in-container tmux client stayed attached after the wrapper was killed, so the
@@ -841,7 +841,7 @@ export function buildForeignDockerAttachCommand(opts: ForeignDockerAttachOptions
* exec process.
*
* Killing a session by name affects ONLY that session, never the group's other
* members — so this can only ever remove the view Codeman created, never the
* members, so this can only ever remove the view Codeman created, never the
* human's session it is grouped with. Sibling of `buildDockerKillCommand`, and
* fired best-effort on the same path.
*/
@@ -873,9 +873,9 @@ export function foreignRemoteShellCommand(script: string): string {
}
export interface ForeignRemoteAttachOptions extends ForeignAttachTarget {
/** `buildSshConnectionArgs(host)` — `['ssh', '-o BatchMode=yes', ...]`. */
/** `buildSshConnectionArgs(host)`: `['ssh', '-o BatchMode=yes', ...]`. */
sshArgs: string[];
/** `remoteSshTarget(host)` — `user@host`. */
/** `remoteSshTarget(host)`: `user@host`. */
sshTarget: string;
}
@@ -883,7 +883,7 @@ export interface ForeignRemoteAttachOptions extends ForeignAttachTarget {
* Remote adoption: ssh in and attach there.
*
* `-t` goes right after `ssh -o BatchMode=yes`, exactly where
* `buildRemoteAttachCommand` puts it — interactive tmux needs a PTY. Connection
* `buildRemoteAttachCommand` puts it: interactive tmux needs a PTY. Connection
* options come from the shared `buildSshConnectionArgs`, never hand-built here,
* so adoption reaches a proxied/custom-port host on the same terms discovery did.
*/