Codecov Report

❌ Patch coverage is 83.72093% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 52.63%. Comparing base (6ca471e) to head (b11c74f).

@@                     Coverage Diff                     @@
##           fix/723-repr-socket-dir     #730      +/-   ##
===========================================================
+ Coverage                    52.40%   52.63%   +0.22%     
===========================================================
  Files                           26       26              
  Lines                         3742     3764      +22     
  Branches                       747      747              
===========================================================
+ Hits                          1961     1981      +20     
- Misses                        1477     1481       +4     
+ Partials                       304      302       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

Code review

Found 1 issue:

1. The socket_name branch validates a socket path libtmux computes, but tmux resolves symlinks before binding, so a configuration tmux handles fine is now refused at construction with no opt-out.

Reproduced on this head, with TMUX_TMPDIR set to a 93-char symlink pointing at a short directory:

$ TMUX_TMPDIR=$LONG tmux -f /dev/null display-message -p '#{socket_path}'
/tmp/sd3394337/tmux-1000/default

tmux realpath()s the socket directory in make_label(), so the real path is 34 bytes and binds without complaint. libtmux's resolve_socket_path() does not, so it measures the unresolved 93-char symlink instead:

SocketPathTooLong: Socket path for socket_name='probe' is 109 bytes, over the 107 byte limit

decee4bf removed this same computation from Server.__init__ and gave the reason in its body — "Guessing the path would have been wrong anyway. tmux resolves a -L name against its own socket directory, which honours TMUX_TMPDIR, the per-uid directory and the platform's overrides. Callers that need the real path ask tmux for it" — and pinned the contract with test_socket_path_not_derived_from_socket_name. #727 reintroduced the guess but kept it harmless: repr text and a best-effort unlink, with a docstring repeating the caveat that the path is "computed, not observed". This PR turns the same guess into a hard failure.

The socket_path is not None branch is sound, since tmux uses -S verbatim with no realpath(). Only the resolved-from-socket_name branch inherits the guess. Options: os.path.realpath() the socket directory before measuring, or keep the eager check for the caller-supplied socket_path and defer the resolved-name case to where the path meets tmux.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

Force-pushed a redesign of where the check runs. The earlier revision validated the socket path in Server.__init__ on every route; two measurements against tmux 3.7b showed that is unsound for the routes that resolve through the environment.

tmux reads $TMUX_TMPDIR at exec, not at construction. Constructing Server(socket_name=N) with TMUX_TMPDIR=a, changing it to b, then dispatching, binds under b. A construction-time check on that route therefore tests a value with no bearing on what tmux uses — it passes when the variable later grows, which is the exact File name too long this issue is about, and refuses when it later shrinks.

A bare client prefers $TMUX over $TMUX_TMPDIR. With $TMUX set, tmux list-sessions succeeds while $TMUX_TMPDIR names a directory far too deep to bind. The previous revision refused to construct a bare Server() there — in the case libtmux is most often used from, a script running inside tmux.

socket_path= is unaffected and is still refused at construction, since libtmux passes it through unchanged as -S<path>.

This also closes the design question raised in review: Server.is_alive() stays total, because a named or bare Server always constructs.

Code review

Found 1 issue, now fixed in 337c921:

1. test_bare_server_inside_a_pane_ignores_tmux_tmpdir patched $TMUX_TMPDIR for the whole test, leaking a live tmux daemon on every run. The server fixture takes monkeypatch, so pytest undoes the patches only after that fixture's finalizer has run — _reap_test_server then resolves the socket under the 120-char directory, cannot reach the server so skips kill(), and unlinks a path that never existed. Both halves of the reap contract fail silently while the test passes.

https://github.com/tmux-python/libtmux/blob/593d5ffd2ba4a1a0e9ac0e1fa06ba4a6e40b93cd/tests/test_server.py#L323-L332

Reproduced deterministically on that revision: the test passes, then tmux -L libtmux_test<...> list-sessions lists two live sessions. This is the same failure mode raised on the parent PR in this stack (#727), and it defeats the invariant _reap_test_server was written for (#660, stale sockets accumulating indefinitely).

Fixed by scoping the patches to the assertions with monkeypatch.context(), matching how #727 resolved it. Verified: no sockets left behind after the test, and none after the full tests/test_server.py run.

Also corrected in the same push: Server.socket_args()'s docstring said the check "belongs at dispatch rather than at construction", which stopped being accurate once an explicit socket_path became a construction-time check. It now states the split.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

Code review

Found 3 issues, all fixed in 66ff46f:

1. libtmux refused servers tmux reaches. tmux creates <dir>/tmux-<euid> and silently binds under /tmp when it cannot, so a $TMUX_TMPDIR that does not exist never holds the socket however long its name is. Measuring it anyway made is_alive() report a live daemon dead and the first command raise.

https://github.com/tmux-python/libtmux/blob/c8356b9d0e5e3f9b4e9c9e2f0e4f3d2a1b6c7e8f/src/libtmux/_internal/env.py#L143-L149

Measured: TMUX_TMPDIR=/tmp/<200 d's> tmux -L probe new-session -d exits 0 with the socket at /tmp/tmux-1000/probe, while libtmux reported is_alive() -> False and raised SocketPathTooLong. resolve_socket_path() now mirrors the fallback; a directory that exists and is too deep still raises.

2. __repr__ named a socket the object does not use. It resolved from $TMUX_TMPDIR even inside a pane, where a bare client uses $TMUX — the rule this PR itself establishes for socket_args().

3. Three tests and a doctest pinned the false refusal by pointing $TMUX_TMPDIR at a nonexistent directory. They now use a real one, so they still exercise the true positive.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

Code review

Found 2 issues, fixed in 57d931d:

1. The previous round's fix was only half applied. Server.is_alive's doctest and the pytest-plugin usage page still pointed $TMUX_TMPDIR at a directory that does not exist, so after the /tmp fallback landed they stopped reaching the length check at all — they passed for the wrong reason.

https://github.com/tmux-python/libtmux/blob/4ad7f9c7e6b3c9a1f0d2e8b7a4c5d6e9f0a1b2c3/src/libtmux/server.py#L367-L372

Both now create the directory first, so the resolved path is one tmux would really bind.

2. The Server class docstring built its directory with tempfile.mkdtemp(), leaking one per run. It now uses the tmp_path fixture through request, which pytest rotates.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

Code review

Found 2 issues, fixed in adcc295:

1. The stated tmux mechanism was wrong in four places. "tmux creates <dir>/tmux-<euid> and quietly falls back to /tmp when it cannot" describes the uncreatable case, which is a hard error rather than a fallback. Measured on tmux 3.7b: a nonexistent directory and a broken symlink both fall back and exit 0, while a directory that exists but is unwritable, or is a regular file, both fail with couldn't create directory. tmux takes the first of $TMUX_TMPDIR and /tmp that resolves.

https://github.com/tmux-python/libtmux/blob/57d931d0e0c9b0a0f8d6e2c1a3b5d7f9e0a2c4b6/src/libtmux/_internal/env.py#L111-L116

2. is_dir() was the wrong predicate for that rule. exists() follows symlinks the way realpath() does, so a broken link falls back as tmux does, and a regular file is left to tmux to reject rather than silently resolving to a /tmp path tmux would never bind. Verified both the original regression and the true positive still behave correctly.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}