davinci-resolve-mcp 4.8.20 → 4.8.22

package/CHANGELOG.md

@@ -2,6 +2,60 @@
 
 Release history for the DaVinci Resolve MCP Server. The latest release is summarized in the root README; older entries live here to keep the README focused.
 
+## What's New in v4.8.22 — the control panel port check cannot hang on a wedged lsof
+
+### Fixed
+
+- **`open_control_panel` could block forever behind an unkillable `lsof`.** The
+  port-owner check ran `lsof -iTCP:<port> -sTCP:LISTEN -t` through
+  `subprocess.run(timeout=3)`. On macOS, lsof wedges in uninterruptible kernel wait
+  (state `U` in `ps`) when a network mount is stale, and a process in that state
+  ignores SIGKILL. `subprocess.run`'s timeout path kills the child and then waits
+  for it, so the 3-second timeout never returned: the caller hung with the child.
+  Measured on 2026-09-26 on the release machine, where 489 lsof processes had been
+  stuck for 12 hours and the offline suite sat in this function for 13 minutes.
+  `_port_owner_pid` now starts lsof in its own session, polls to the deadline, and
+  on expiry sends SIGKILL and abandons the child instead of joining it. stdout is
+  read only once `poll()` reports an exit, and the pipe is closed on every path.
+  A missing lsof is still `None`, not an exception.
+
+### Tests
+
+- `tests/test_port_owner_pid.py`: a fake child whose `poll()` never returns and
+  whose `kill()` is a no-op yields `None` within the deadline with `wait()` and
+  `communicate()` never called and the pipe closed; an exited child still yields
+  its PID; a missing binary yields `None`; and a real subprocess that ignores
+  SIGTERM, with `kill` patched out, is left running rather than joined.
+
+## What's New in v4.8.21 — timeline duration no longer overcounts by one frame
+
+### Fixed
+
+- **Two timeline-duration readers reported one frame too many.** ([#269](https://github.com/samuelgursky/davinci-resolve-mcp/pull/269), @Dev-next-gen)
+  The granular `get_current_timeline` and `get_project_info`'s per-timeline
+  `duration` computed `GetEndFrame() - GetStartFrame() + 1`, while the compound
+  server, `brain_edits` and `render_stress` all use `GetEndFrame() - GetStartFrame()`
+  for the same timeline. `GetEndFrame()` is one past the last frame, so the `+ 1`
+  counted a frame that does not exist: a 600-frame timeline read back as 601. Both
+  readers now agree with the rest of the server.
+
+### Documentation
+
+- **`api_truth` records `Timeline.GetEndFrame` as an exclusive bound.** Measured
+  live on Studio 19.1.3.7 while reviewing #269: a timeline whose only full-length
+  item is a 32742-frame clip reads start 0 / end 32742, and the item's
+  `GetEnd()` is also 32742. The entry sits beside the existing
+  `AppendToTimeline clipInfo endFrame` entry, which documents the same half-open
+  convention on the write side, so the frame-count rule is now findable from
+  either direction.
+
+### Tests
+
+- `tests/test_timeline_duration.py` (from the PR): both readers return
+  `end - start` for a stubbed 86400..87000 timeline; fails on v4.8.20 with 601.
+- `tests/test_api_truth.py`: the new ledger entry is findable by `GetEndFrame`,
+  carries the `off-by-one` tag, and states the `end - start` rule.
+
 ## What's New in v4.8.20 — every boolean param honours `"false"`
 
 ### Fixed

package/README.md

@@ -2,7 +2,7 @@
 
 English | [简体中文](README.zh-CN.md)
 
-[![Version](https://img.shields.io/badge/version-4.8.20-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
+[![Version](https://img.shields.io/badge/version-4.8.22-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
 [![npm](https://img.shields.io/npm/v/davinci-resolve-mcp.svg?label=npm&color=CB3837)](https://www.npmjs.com/package/davinci-resolve-mcp)
 [![API Coverage](https://img.shields.io/badge/API%20Coverage-100%25-brightgreen.svg)](docs/reference/api-coverage.md)
 [![Tools](https://img.shields.io/badge/MCP%20Tools-37%20(389%20full)-blue.svg)](#server-modes)

package/README.zh-CN.md

@@ -2,7 +2,7 @@
 
 [English](README.md) | 简体中文
 
-[![Version](https://img.shields.io/badge/version-4.8.20-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
+[![Version](https://img.shields.io/badge/version-4.8.22-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
 [![npm](https://img.shields.io/npm/v/davinci-resolve-mcp.svg?label=npm&color=CB3837)](https://www.npmjs.com/package/davinci-resolve-mcp)
 [![API Coverage](https://img.shields.io/badge/API%20Coverage-100%25-brightgreen.svg)](docs/reference/api-coverage.md)
 [![Tools](https://img.shields.io/badge/MCP%20Tools-37%20(389%20full)-blue.svg)](#服务器模式)
@@ -12,7 +12,7 @@
 [![Python](https://img.shields.io/badge/python-3.10+-green.svg)](https://www.python.org/downloads/)
 [![License](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 
-> 本翻译对应 v4.8.20 版 README。如与英文原版有出入，以 [英文原版](README.md) 为准。
+> 本翻译对应 v4.8.22 版 README。如与英文原版有出入，以 [英文原版](README.md) 为准。
 
 一个 Model Context Protocol (MCP) 服务器，让 AI 助手通过官方脚本 API 控制 DaVinci Resolve Studio（达芬奇）。它提供完整的 API 覆盖，外加带护栏的工作流助手，涵盖剪辑、媒体池整理、渲染设置、审阅标记、调色、Fusion、Fairlight、项目生命周期任务、扩展开发，以及不碰源媒体的媒体分析。
 

package/install.py

@@ -37,7 +37,7 @@ from src.utils.update_check import (
 
 # ─── Version ──────────────────────────────────────────────────────────────────
 
-VERSION = "4.8.20"
+VERSION = "4.8.22"
 # Only hard floor: mcp[cli] requires Python 3.10+. There is no upper bound —
 # Resolve's scripting bridge loads into newer interpreters on recent builds
 # (Python 3.14 verified against Resolve Studio 20.3.2). Older Resolve builds

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "davinci-resolve-mcp",
-  "version": "4.8.20",
+  "version": "4.8.22",
   "description": "NPM bootstrapper for the DaVinci Resolve MCP Server.",
   "license": "MIT",
   "author": "Samuel Gursky <samgursky@gmail.com>",

package/src/granular/common.py

@@ -93,7 +93,7 @@ if not logging.getLogger().handlers:
         handlers=[logging.StreamHandler()],
     )
 
-VERSION = "4.8.20"
+VERSION = "4.8.22"
 logger = logging.getLogger("davinci-resolve-mcp")
 logger.info(f"Starting DaVinci Resolve MCP Server v{VERSION}")
 logger.info(f"Detected platform: {get_platform()}")

package/src/granular/timeline.py

@@ -65,7 +65,7 @@ def get_current_timeline() -> Dict[str, Any]:
             "width": current_timeline.GetSetting("timelineResolutionWidth"),
             "height": current_timeline.GetSetting("timelineResolutionHeight")
         },
-        "duration": current_timeline.GetEndFrame() - current_timeline.GetStartFrame() + 1
+        "duration": current_timeline.GetEndFrame() - current_timeline.GetStartFrame()
     }
     
     return result

package/src/server.py

@@ -11,7 +11,7 @@ Usage:
     python src/server.py --full       # Start the 377-tool granular server instead
 """
 
-VERSION = "4.8.20"
+VERSION = "4.8.22"
 
 import base64
 import os
@@ -17750,24 +17750,57 @@ def _control_panel_remote_version(host: str, port: int, timeout: float = 1.5) ->
     return _control_panel_probe(host, port, timeout).get("version")
 
 
-def _port_owner_pid(host: str, port: int) -> Optional[int]:
+def _port_owner_pid(host: str, port: int, timeout: float = 3.0) -> Optional[int]:
     """Return PID of the process LISTENing on `port`, or None if free/unknown.
 
     Uses lsof with `-iTCP:<port> -sTCP:LISTEN -t`: one PID per line, no header.
     Host is informational only — lsof matches any local LISTEN socket on that
     port (which is what we care about for port-collision detection).
+
+    The deadline is enforced by polling, not by ``subprocess.run(timeout=)``.
+    On macOS, lsof can wedge in uninterruptible kernel wait (state ``U`` in
+    ``ps``) when a network mount is stale, and a child in that state ignores
+    SIGKILL. ``subprocess.run``'s timeout path kills the child and then WAITS
+    for it, so the caller hung with it — measured 2026-09-26 on the release
+    machine: 489 lsof processes stuck in ``U`` for 12 h, and the offline suite
+    blocked here for 13 min. Now the child is killed on expiry and abandoned
+    rather than joined; stdout is read only once ``poll()`` says it exited.
     """
     import subprocess
+    import time as _t
     try:
-        result = subprocess.run(
+        proc = subprocess.Popen(
             ["lsof", "-nP", "-iTCP:" + str(port), "-sTCP:LISTEN", "-t"],
-            capture_output=True, timeout=3, text=True, encoding="utf-8",
-            errors="replace", check=False,
-            stdin=subprocess.DEVNULL,
+            stdout=subprocess.PIPE, stderr=subprocess.DEVNULL,
+            stdin=subprocess.DEVNULL, start_new_session=True,
         )
-    except (OSError, subprocess.TimeoutExpired):
+    except OSError:
         return None
-    for line in (result.stdout or "").splitlines():
+    deadline = _t.monotonic() + max(0.0, float(timeout))
+    out = b""
+    try:
+        while proc.poll() is None:
+            if _t.monotonic() >= deadline:
+                try:
+                    proc.kill()
+                except OSError:
+                    pass
+                # Deliberately no wait()/communicate(): a wedged lsof never
+                # exits, and joining it is exactly the hang this guards against.
+                return None
+            _t.sleep(0.05)
+        if proc.stdout is not None:
+            try:
+                out = proc.stdout.read() or b""
+            except OSError:
+                out = b""
+    finally:
+        if proc.stdout is not None:
+            try:
+                proc.stdout.close()
+            except OSError:
+                pass
+    for line in out.decode("utf-8", "replace").splitlines():
         line = line.strip()
         if line.isdigit():
             return int(line)

package/src/utils/api_truth.py

@@ -2005,6 +2005,30 @@ API_TRUTH: List[Dict[str, Any]] = [
                        "when mirroring keep-ranges into clipInfos.",
         "tags": ["timeline", "edit", "off-by-one", "readback"],
     },
+    {
+        "symbol": "Timeline.GetEndFrame (exclusive bound)",
+        "object": "Timeline",
+        "signature": "() -> int",
+        "reality": "GetEndFrame() is an EXCLUSIVE bound, one past the last "
+                   "frame: a timeline's frame count is "
+                   "GetEndFrame() - GetStartFrame(), not that plus one. "
+                   "Measured live on Studio 19.1.3.7 (2026-09-26): a "
+                   "timeline whose only full-length item is a 32742-frame "
+                   "clip (media-pool 'Frames' 32742, 'End' 32741) reads "
+                   "GetStartFrame() 0 / GetEndFrame() 32742, and that item's "
+                   "TimelineItem.GetEnd() is also 32742 - the same half-open "
+                   "convention as clipInfo endFrame above. Until v4.8.21 two "
+                   "readers here (granular get_current_timeline and "
+                   "get_project_info's per-timeline duration) added 1 and "
+                   "reported 601 frames for a 600-frame timeline while every "
+                   "other duration path in the server did not (PR #269).",
+        "recommended": "duration_frames = GetEndFrame() - GetStartFrame(). "
+                       "The last frame index is GetEndFrame() - 1. Apply the "
+                       "same rule to TimelineItem.GetEnd(): an item occupies "
+                       "[GetStart(), GetEnd()).",
+        "tags": ["timeline", "off-by-one", "readback", "duration"],
+        "verified_on": "DaVinci Resolve Studio 19.1.3.7",
+    },
     {
         "symbol": "MediaPool.ImportTimelineFromFile (internal sequence name overrides timelineName)",
         "object": "MediaPool",

package/src/utils/project_properties.py

@@ -592,7 +592,7 @@ def get_project_info(project_obj) -> Dict[str, Any]:
                 timeline_info = {
                     "name": timeline.GetName(),
                     "isCurrent": timeline.GetName() == current_timeline_name,
-                    "duration": timeline.GetEndFrame() - timeline.GetStartFrame() + 1
+                    "duration": timeline.GetEndFrame() - timeline.GetStartFrame()
                 }
                 project_info["timelines"].append(timeline_info)