Skip to content

CHANGELOG.md

Latest commit

 

History

History
259 lines (209 loc) · 11.6 KB

CHANGELOG.md

File metadata and controls

259 lines (209 loc) · 11.6 KB

Changelog

All notable changes to this project will be documented in this file.

[0.2.5] - 2026-04-01

Features

  • Auto-push tag when release PR is merged リリースPR（release/v*ブランチ）がmainにマージされると GitHub Appトークンで自動的にタグをpushし、releaseワークフローをトリガーする

Refactor

  • Merge tag-release into release workflow tag-releaseとreleaseを統合し、リリースPRマージ時に タグ作成とGitHub Release作成を1つのワークフローで完結させる

[0.2.4] - 2026-04-01

Bug Fixes

  • Quote colon-containing strings in release-prep workflow YAML requires quoting values that contain colons. The title and commit-message fields were causing a parse error on line 38.
  • Use GitHub App token for release prep PR creation
  • Add actions/create-github-app-token@v2 to generate token
    • Update peter-evans/create-pull-request v7 -> v8 (Node.js 24対応)
    • PRをGitHub Appとして作成することでCIワークフローがトリガーされる
  • Allow ttak0422-bot in claude code review workflow

[0.2.3] - 2026-03-28

Features

  • Add nix app for testing pterm in isolated neovim Add nix run .#test-nvim app that launches Neovim with pterm pre-loaded for quick manual verification without manual setup.

Bug Fixes

  • Update vt100 0.15 → 0.16 to restore dim (SGR 2) attribute in snapshots (#16) vt100 0.15 did not track the dim/faint attribute (SGR 2) at all — no TEXT_MODE_DIM bit and no handler in sgr(). As a result, state_formatted() never re-emitted ESC[2m, so text rendered as dim in the live session would appear as normal-weight text after a detach/re-attach cycle.

    vt100 0.16 adds TEXT_MODE_DIM and handles SGR 2 → set_dim(), and state_formatted() correctly re-emits Intensity::Dim as ESC[2m.

    Confirmed via byte-level comparison of live PTY output vs snapshot: ORIGINAL: ESC[2mNo ESC[0m (faint) 0.15 snap: ESC[m No (dim lost) 0.16 snap: ESC[m ESC[2mNo (dim preserved)

    API changes in 0.16:

    • screen.errors() removed → dropped from debug log format string
    • parser.set_size(rows, cols) → parser.screen_mut().set_size(rows, cols)

  • Improve session snapshot and terminal lifecycle management (#17)

  • fix: improve session snapshot and lifecycle management

    • (H) Send terminal cleanup sequences on detach to reset mouse modes, bracketed paste, alternate screen, and cursor visibility in the client terminal
    • (B) Increase scrollback capacity from 0 to 10,000 lines so history is available after reattach
    • (G) Respond to DA1/DA2 queries with canned responses when no clients are connected, preventing apps from hanging on capability detection
    • (A) Prepend ESC[?1049h in snapshots when alternate screen is active so reattaching clients properly enter alternate screen mode
    • (C) Track window title via vt100 Callbacks and restore it in snapshots via OSC 2 sequence on reattach

[0.2.2] - 2026-03-11

Bug Fixes

  • Defer BufDelete handler to ignore buflisted changes from scope-nvim scope-nvim scopes buffers per tab page by toggling buflisted on TabLeave/TabEnter. Setting buflisted=false fires BufDelete, which pterm's handler interpreted as a true buffer deletion—calling M.detach(), killing the bridge (SIGHUP/129), and wiping the buffer.

    Defer the BufDelete callback with vim.schedule and check whether the buffer is still valid and loaded. A merely-unlisted buffer survives the check and the connection is preserved; a truly deleted buffer triggers detach as before.

[0.2.1] - 2026-03-10

Bug Fixes

  • Prevent duplicate rendering when attaching via :Pterm user command When a client with pending_snapshot=true received a snapshot triggered by PTY output arrival, the same flush cycle also broadcast the raw OUTPUT bytes to that client. Since the snapshot already reflects the effect of those bytes (read_pty feeds data to the VT parser before flush), Neovim's libvterm processed the same content twice, causing garbled/duplicated rendering (e.g. "e" → "ee" → "eechooo").

    This was not observed with toggleterm because pterm open connects the bridge in the same process immediately after daemon creation, before significant PTY output is produced. The :Pterm user command creates the daemon separately (pterm new via vim.fn.system), so by the time the bridge attaches the shell prompt is already in the VT state, making the race between snapshot and OUTPUT much more likely.

  • Use pterm open to eliminate snapshot timing gap in :Pterm command When :Pterm created a session via the two-step flow (vim.fn.system for pterm new → wait_for_socket → jobstart for pterm attach), the shell had time to produce PTY output during the gap. On bridge connection, server.rs flush_pty_output() triggered snapshot delivery at the old 80×24 size BEFORE the client's RESIZE was processed, leaving libvterm with wrong scroll regions and causing progressive display corruption.

    CLI pterm open was unaffected because it handles creation and bridge attachment in a single process with minimal delay.

Miscellaneous Tasks

  • Update CHANGELOG.md on release Add post-release steps to regenerate CHANGELOG.md using git-cliff and commit it back to the default branch.

[0.2.0] - 2026-03-08

Features

  • (server) Handle redraw request
  • (cli) Add redraw command
  • (nvim) Add PtermRedraw command
  • Add Telescope extension for pterm sessions Add lua/telescope/_extensions/pterm.lua providing a Telescope picker (:Telescope pterm sessions) to fuzzy-search and jump to pterm sessions. Connected sessions are visually marked as [connected]. Selecting a stale session shows an error notification without crashing. Also auto-loads the extension from M.setup() when Telescope is available.

Bug Fixes

  • (server) Improve error handling and flush after redraw broadcast Gracefully handle client I/O errors in the event loop instead of propagating them, and ensure all clients are flushed after a redraw broadcast so the snapshot is delivered immediately.

  • (vim) Stabilize pterm session completion

  • (server) Coalesce PTY output into single message per poll cycle Set the PTY master fd to non-blocking after openpty() and add a drain loop in handle_pty_output that reads all available data before sending a single OUTPUT message to clients. This prevents fragmented prompt rendering (e.g. "/path ma" → "/path mai" → "~/path main") caused by forwarding each small PTY read as a separate protocol message.

    • pty.rs: set O_NONBLOCK on master fd
    • session.rs: map EAGAIN/EWOULDBLOCK to io::ErrorKind::WouldBlock
    • server.rs: drain loop aggregates all reads into one OUTPUT message

  • (bridge) Batch OUTPUT writes per poll cycle Collect all OUTPUT/SCROLLBACK payloads from protocol frames into a single buffer and write once to stdout, instead of writing each frame individually. This further reduces rendering fragmentation on the client side.

  • (server) Add bounded micro-batching for PTY output coalescing Introduce tmux-style micro-batching to prevent prompt fragmentation on fast machines where poll→read outruns shell writes. Small PTY outputs are held up to 1ms (max 3ms burst) before flushing, while large outputs (≥4096 bytes) flush immediately.

    Also fixes:

    • EXIT message ordering: queue into send_buf instead of direct write_all to preserve OUTPUT→EXIT ordering under backpressure
    • Prevent duplicate EXIT broadcast via exit_sent flag
    • Enforce size bound during drain loop (mid-loop flush)
    • Gate deadline refresh on actual new reads

  • (server) Avoid OUTPUT before initial snapshot

  • (server) Replace timer-based batching with drain-and-flush Remove BATCH_DELAY_MS/BATCH_MAX_MS/BATCH_FLUSH_SIZE micro-batching and SNAPSHOT_DEFER_MS (500ms) timer. These caused noticeable UX latency.

    New approach:

    • Drain PTY non-blocking until WouldBlock, then flush immediately
    • Snapshot sent on RESIZE or on first PTY OUTPUT (no timer deferral)
    • Bridge-side per-poll-cycle batching retained (zero-delay)

Refactor

  • Remove auto load_extension from setup Follow the standard Telescope extension convention where users explicitly call require("telescope").load_extension("pterm") instead of auto-loading during setup. This aligns with how all major Telescope extensions (fzf-native, file-browser, frecency, undo, ui-select) work and better supports lazy-loading scenarios.

Documentation

  • Rewrite README for Nix users and move detailed config to docs/
  • Replace lazy.nvim install section with Nix flake-based instructions
    • Remove verbose sections (Build, How It Works, Session Lifecycle, etc.)
    • Add default configuration example under setup()
    • Move environment variables and socket location to docs/configuration.md
  • Update DESIGN.md and proto comments to reflect vt100 state tracking
  • Replace "raw byte scrollback ring" with vt100 parser-based terminal state tracking in DESIGN.md architecture diagram and daemon section
    • Update SCROLLBACK message description to reflect state_formatted snapshots
    • Remove incorrect proto comment about omitting length field for empty payloads
    • Add trailing newlines to DESIGN.md and proto/src/lib.rs
  • Update README to reflect current implementation Document CLI options (--cols, --rows, -- ), hierarchical session names, list prefix filter, default session name, exported Lua API functions, and clarify cols/rows as fallback values.
  • Update DESIGN.md with output coalescing and batching details
  • Add Telescope extension section to README

Styling

  • Format bridge and main

Miscellaneous Tasks

  • Add release workflow for tag-triggered changelog generation
  • (docs) Rename configuration.md to CONFIGURATION.md

[0.1.0] - 2026-02-26

Features

  • Initial commit

  • (core) Add open command to create missing session before attach

  • (core) Replace raw scrollback with vt100 terminal state tracking Use the vt100 crate to maintain full terminal emulation state instead of a raw byte ring buffer. On client attach, generate a precise state snapshot via state_formatted() that correctly reproduces cursor position, visibility, SGR attributes, scroll region, and input modes.

    This supersedes the previous scrollback sanitization and post-replay reset heuristics (Plan B), fixing cursor disappearance and color corruption when re-attaching to sessions.

Bug Fixes

  • (core) Write all input bytes to pty to avoid garbled redraw
  • (core) Sanitize scrollback replay to drop terminal queries
  • (core) Reset terminal state after scrollback replay Scrollback may contain stale SGR attributes or cursor-hide sequences left by applications (vim, fzf, etc.). Without a reset, re-attaching clients could see invisible cursors or garbled colors (e.g. zsh autosuggestions blending into the background). Append SGR 0 and DECCM show-cursor after the scrollback payload to ensure a clean terminal state for the live session.

Documentation

  • (core) Use actual repository path in lazy.nvim example
  • (core) Remove cachix section from readme

Miscellaneous Tasks