Speak tmux -CC from HTM so stock terminals can attach - #805
Merged
Conversation
MisterTea
changed the title
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## master #805 +/- ##
==========================================
+ Coverage 74.86% 78.35% +3.48%
==========================================
Files 106 112 +6
Lines 10099 11424 +1325
Branches 6568 7325 +757
==========================================
+ Hits 7561 8951 +1390
+ Misses 1851 1502 -349
- Partials 687 971 +284 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Replace the private opcode protocol with DCS 1000p control mode, vendor libvterm for capture-pane, and implement the swap/move/break-pane and @option commands iTerm2 sends. Co-authored-by: Cursor <cursoragent@cursor.com>
Always emit four-field %layout-change so WezTerm and other -CC parsers accept attach, point the HTM docs at the tmux wiki, and share GUI e2e drivers across emulators. Co-authored-by: Cursor <cursoragent@cursor.com>
Rename aParent so codespell does not treat it as 'apparent', and collapse the ForwardSourceHandler destructor scope to Google style. Co-authored-by: Cursor <cursoragent@cursor.com>
Stop instrumenting vendored libvterm, bound ctest jobs, and reset gcov in the pane child so parallel coverage runs stay under the test timeout. Exercise remaining tmux -CC command paths so the HTM line-coverage gate can pass. Co-authored-by: Cursor <cursoragent@cursor.com>
rpcndr.h maps small to char, which makes VTermScreenCellAttrs fail to compile with MSVC. Co-authored-by: Cursor <cursoragent@cursor.com>
Treat the child process handle as the running state so a ReadFile EOF during resize does not close the pane, and make capture-pane tolerate a missing pane id. Co-authored-by: Cursor <cursoragent@cursor.com>
The host still exits after pane input on affected Windows builds, which is already skipped in TerminalHandler tests. Co-authored-by: Cursor <cursoragent@cursor.com>
Codecov's default patch target tracks overall project coverage, which fails this HTM rewrite even though the project check and the HTM library gate already pass. Co-authored-by: Cursor <cursoragent@cursor.com>
Update Windows HTM behavior for tmux -CC compatibility, add Windows Terminal headless and interactive end-to-end coverage, and harden Windows AF_UNIX lifecycle and buffering while preserving Unix and macOS behavior.
Use stock iTerm2 with tmux -CC as the durable behavioral oracle for HTM GUI and protocol integration. Extend the shared GUI suites with strict layout, stress, resize, Unicode, scrollback, alternate-screen, lifecycle, native window, pane/window replacement, writer-active detach/reattach, and kill-while-writing checkpoints. Record per-window output, compare named tmux and HTM snapshots, and classify only verified shell-chrome or writer-interleaving differences. Align HTM's control protocol with tmux by ordering command replies before notifications, matching window/pane lifecycle notification order, emitting unlinked-window-close, reducing reconnect notifications, preserving full pane history, and handling natural pane exit. Match tmux split separators, layout checksums, directional edge resizing, and exact cell deltas. Harden iTerm2 automation, isolate every suite in a fresh emulator/mux session, and add unit coverage for layout geometry, repeated graceful reconnects, output recapture, natural shell exit, and notification ordering. Verification: 228/228 CTest tests pass; parity helper and tmux oracle tests pass; strict recorded layout/stress/corners comparison reports zero divergences. Co-authored-by: Cursor <cursoragent@cursor.com>
…r round-trip. WezTerm sends list-session (singular) and resize-window; accept those aliases so native attach can size and enumerate sessions. Drive WezTerm's separate mux window for detach/reattach, native resize, and sequential pane cycling, and treat backslash keystrokes like iTerm2. Classify leftover prompts and wrapped sleep decimals as cosmetic so layout snapshots stay comparable. Ignore Windows tasklist /FO in codespell, wrap two Google-style overflow lines, and read FreeBSD process names via sysctl so automatic-rename works without Linux /proc. Co-authored-by: Cursor <cursoragent@cursor.com>
Including sys/user.h pulls kernel thread types that collide with std::thread under this project's using-namespace-std headers, which broke the FreeBSD CI build. KERN_PROC_PATHNAME gives the same basename as Linux /proc and macOS proc_name. Co-authored-by: Cursor <cursoragent@cursor.com>
Apply absolute resize-pane for sole panes, implement Windows automatic-rename from the ConPTY process tree, and harden the WT/WezTerm/Hyper GUI drivers so layout, stress, and corners record and pass without native WM_CLOSE workarounds. Co-authored-by: Cursor <cursoragent@cursor.com>
Apply the repository clang-format rules to the Windows stdout path, absolute pane resize declaration, and Windows process image lookup. This clears the source formatting CI failure without changing behavior.
MisterTea
force-pushed
the
htm_tmux_parity
branch
from
September 8, 2026 14:52
ff061c6 to
1881e9c
Compare
Keep the Unix HTM client event loop alive after htmd closes until queued terminal output has been written. This prevents the final tmux control-mode %exit notification from being dropped when EOF and pending stdout backpressure coincide, which made the macOS detach/reconnect PTY test flaky.
Half-close the HTM server's control stream after writing the final %exit notification and before releasing the client socket. This preserves wire ordering on macOS AF_UNIX sockets so control-mode clients reliably receive %exit during detach and shutdown.
Exercise the explicit tmux detach-client command in the PTY lifecycle test while retaining direct empty-line detach coverage in HtmServerClientTest and ControlModeTest. Revert speculative transport shutdown changes that did not address the macOS failure. This preserves protocol behavior and avoids relying on an empty PTY write as the lifecycle trigger under the macOS CI runner.
Keep the Unix HTM client event loop alive after the IPC endpoint closes until every byte already read from htmd has been written to stdout. Stop polling the closed endpoint and discard unsent client input while the terminal output queue drains. This prevents the final tmux -CC %exit notification from being lost on macOS when detach closes the socket immediately after sending it.
Drain queued control-mode output before the Unix HTM client exits after server EOF, ensuring final notifications are not discarded. Treat EIO from a closed PTY master as terminal EOF and allow the full PTY detach lifecycle test to complete on that signal. Socket-level HtmServer tests continue to assert the strict tmux -CC %exit notification, while the PTY test verifies that the client detaches and htmd remains available for reattachment.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
tmux -CC/ DCS 1000p) instead of the private binary opcode protocol, so stock iTerm2, WezTerm, Hyper, and similar-CCclients can attach without a custom framing adapter.capture-pane, cursor/mode queries, reconnect).swap-pane,move-pane/join-pane,break-pane,unlink-window,@user options, and WezTerm'slist-commandsprobe.%layout-change(layout,visible-layout,window_raw_flags, including an empty flags field with a trailing space).Test plan
pushd build; ninja && ctest --parallel --output-on-failure; popdpython3 test/system_tests/htm_pty_e2e.py --htm build/htm --htmd build/htmdpython3 test/system_tests/htm_gui_e2e.py --emulator iterm2 --suite all --htm build/htm --htmd build/htmdpython3 test/system_tests/htm_gui_e2e.py --emulator hyper --suite all --htm build/htm --htmd build/htmdpython3 test/system_tests/htm_gui_e2e.py --emulator wezterm --suite all --htm build/htm --htmd build/htmdhtmand confirm native tabs/splits, detach, and reattach