Root cause:
The API socket is bound before App evaluates session restore. Waiting only
for that socket lets the late-session test inject its file during startup,
so the server may load it normally instead of exercising late-file recovery.
The missing backup then fails an otherwise correct persistence assertion.
Fix:
Wait for a read-only workspace-list request to pass through App and assert
that the workspace list is still empty before injecting the session file.
Keep the existing autosave deadline and byte-for-byte backup assertion.
Add no sleeps, retries, longer timeouts or production changes.
Validation:
The original assertion fails with unchanged master. With the barrier, five
independent runs pass. Full native just ci passes 3624 Rust tests with six
default skips, formatting, all-target Clippy, 112 maintenance tests, six
architecture tests and 39 integration-asset tests. All seven docs contract
tests pass. Local Windows cross-lint was not run because its SDK is absent.
Co-authored-by: JJ Liebig <jonathan.liebig@gmail.com>
Root cause:
The handoff test starts an agent immediately after workspace creation,
before the new shell is necessarily ready. The reconnect test assumes a
fixed sidebar row and searches raw ANSI output for text that can arrive
as separate incremental cell updates.
Fix:
Wait for a shell builtin to write a readiness marker before sending the
single raw agent.start request. Exercise the gap with a delayed shell.
Locate the recovered workspace in the rendered sidebar and check the
reconstructed screen for selection and input, including overlay borders.
Do not extend existing timeouts or add retries that hide failures.
Validation:
All 25 affected group and helper tests pass. Ten repeated rounds of the
two runtime tests and the row helper pass, for 30 test executions without
retries. Standalone native just ci passes 3582 Rust tests with six default
skips, formatting, all-target Clippy, 106 maintenance tests, six architecture
tests and 39 integration-asset tests. Windows cross-lint was not run because
its SDK is not installed.
Co-authored-by: JJ Liebig <jonathan.liebig@gmail.com>
Live handoff refused any session with more than 64 panes. The pane count
was checked twice against MAX_FDS_PER_HANDOFF, and the transfer itself put
every pane's pty master into one SCM_RIGHTS control message, so the guard
was the only thing keeping the send inside the kernel's per-message limit.
A session past the limit could only be updated by closing panes or by a
normal restart, which ends every pane process.
Send the descriptors in batches of 64 instead and drop both guards. The
receiving side accumulates across recvmsg calls until the expected count
arrives, bounds every SCM_RIGHTS payload it reads by the control bytes the
kernel returned, rejects a batch that carries more descriptors than it
asked for, and closes the descriptors it already holds on any failure. A
session of 64 panes or fewer still produces one batch, so the bytes on the
wire are unchanged and HANDOFF_VERSION stays at 1.
refs #3393
Co-authored-by: kataokatsuki <kataokatsuki@users.noreply.github.com>
Co-authored-by: JJ Liebig <jonathan.liebig@gmail.com>
* fix: run the full applicable test suite on windows
* test: remove redundant test shell assignment
* test: make endpoint and metadata fixtures deterministic
* test: tolerate exited servers during windows probe cleanup
* feat: keep the outer terminal window title in sync with the session
Herdr emulates the terminals in its panes, so an OSC 0/OSC 2 title
written inside a pane stops at Herdr and never reaches the terminal
Herdr itself runs in. That outer title is what window managers read for
title bars, tab bars, and group bars, so it kept showing whatever the
shell or ssh happened to leave behind.
Add `ui.window_title`, rendered from {hostname}, {workspace}, {tab},
{pane}, and {terminal_title}, and push it to the foreground client
whenever it changes. It renders on the server, so {hostname} names the
machine the panes actually run on rather than the machine a thin remote
client runs on, and it is gated on a pending render so an idle loop
never pays for it.
A title is only remembered as delivered once a foreground client takes
it, so the first client to attach is written to rather than skipped.
`client.window_title.set` still wins over the configured title, and
clearing it now hands the title back to `ui.window_title` instead of
only "herdr".
* fix: deliver the outer window title to a newly attached client
ClientConnected assigns the foreground client directly rather than going
through promote_client_to_foreground, so clearing the sent-title cache
there missed the case that matters most: attaching a second terminal to a
running session. The title was usually unchanged, so the sync returned
early and the new terminal kept whatever its shell or ssh had left.
Key the cache on the client that received the title instead of relying on
every foreground assignment to invalidate it.
* perf: keep hidden pane output off the window title path
Output from a hidden or background pane sets needs_render without setting
needs_full_render, and the retained render plan then skips presentation
for it entirely. Syncing the title on needs_render meant every coalesced
hidden-output tick still formatted, sanitized, and allocated a title that
could not have changed, against the hidden-source early exit AGENTS.md
requires.
Every input to the title is app state, which always requests a full
render, so gate on that instead. The one exception is the focused pane's
own terminal title, which arrives through PTY parsing, so ask for a full
render when that changes and the configured title uses it.
* fix: only cache a window title a client writer received
* fix: carry an api set window title across a live handoff
* fix: make outer window title updates event-driven
---------
Co-authored-by: Ogulcan Celik <ogulcancelik@gmail.com>
Can Celik
Eric YueandJJ Liebig
akbashandakbash-bot
caner-akca
David Heinemeier HanssonandOgulcan Celik