Fix Windows compatibility in helpers

[yamarames]

What

Four defects that make the helpers unusable on Windows. Each was found by running the tool against real footage during a fresh install, not by inspection.

1. grade.py — auto-grade crashes on every clip

metadata=print:file=<path> is a filter argument, where : separates options and \ escapes. An absolute Windows path is destroyed by the filter parser:

[AVFilterGraph] No option name near 'UsersUserAppDataLocalTempt1.txt'
subprocess.CalledProcessError: Command '[... 'metadata=print:file=C:\Users\...']' returned non-zero exit status 4294967274

Fixed by passing a bare filename with cwd set, which requires no filter-arg escaping on any platform.

This is the widest-reaching of the four: auto is the default grade mode, and render.py calls auto_grade_for_clip for every segment whenever the EDL says "grade": "auto".

2. render.py — subtitle burn-in fails

Same root cause. The existing escaping handled : but left \ for the filter parser to consume, so libass received a path with no separators:

[AVFilterGraph] Error initializing filters
Error : No such file or directory

Fixed by converting to forward slashes (accepted by ffmpeg on Windows, no-op on POSIX) before escaping the drive colon. Subtitles-applied-last is Hard Rule 1 in SKILL.md, so this hits on the first real subtitled render.

3. Five helpers — UnicodeEncodeError on cp1252 consoles

Windows consoles default to cp1252, which cannot encode the → ≥ ± × … characters used in progress output:

UnicodeEncodeError: 'charmap' codec can't encode character '\u2192' in position 4

render.py alone has 14 such prints, so a long render dies partway with the encode work already spent. Each entry point now reconfigures stdout/stderr to UTF-8.

4. timeline_view.py — unreadable labels

FONT_CANDIDATES listed only macOS and Linux paths, so PIL fell back to a fixed-size bitmap font that ignores the requested size. Word labels and the time ruler were illegible, which defeats the purpose of the drill-down tool at cut-decision time. Added Consolas, Courier New, Segoe UI and Arial after the existing entries, so macOS and Linux resolution order is unchanged.

Verification

Windows 10, ffmpeg 8.1, Python 3.11. A two-segment EDL rendered end to end: per-segment auto-grade → lossless concat → burned subtitles → two-pass loudnorm. A timeline_view of the output confirms both caption cues are visible and the cut lands where the EDL says. Duration matched expectation (4.10s vs 4.0s, encoder rounding).

Escaping alternatives were tested rather than assumed — for the metadata path, forward-slash-plus-single-escape and double-escape both still fail, while quad-escape and bare-filename-plus-cwd both work; the latter was chosen as the one that needs no platform branching.

Risk

No behavior change on macOS or Linux. The cwd change is platform-neutral, .replace("\\", "/") is a no-op on POSIX paths, reconfigure() is guarded by hasattr, and the font additions are appended below the existing candidates.

🤖 Generated with Claude Code

---

Summary by cubic

Fixes four Windows-only bugs in the helper scripts so auto-grade, subtitle burn-in, and timeline labels work end to end on Windows. No behavior change on macOS or Linux.

• Bug Fixes
  • Auto-grade: pass a bare filename and set cwd for metadata=print:file=... to avoid Windows filter-arg escaping crashes.
  • Subtitles: convert paths to forward slashes and escape the drive colon before applying the subtitles filter.
  • Console output: reconfigure stdout/stderr to UTF-8 in all entry points to prevent UnicodeEncodeError on cp1252 consoles.
  • Timeline fonts: add Windows fallbacks (Consolas, Courier New, Segoe UI, Arial) so labels render at the correct size.

^{Written for commit a1b09af. Summary will update on new commits.}

[cubic-dev-ai]

1 issue found across 5 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="helpers/grade.py">

<violation number="1" location="helpers/grade.py:127">
P2: The new `cwd=str(meta_file.parent)` on the sampling subprocess changes the directory the ffmpeg `-i` input is resolved against. `video` is passed straight through from the caller (grade.py standalone passes the raw, possibly-relative `args.input`/`args.analyze` with no `.resolve()`), so a relative input like `python helpers/grade.py input.mp4 -o out.mp4` — a documented usage — will now fail with CalledProcessError because ffmpeg looks for `input.mp4` inside the OS temp dir. This also violates the stated "no behavior change on macOS/Linux" goal since it breaks previously-working relative paths on all platforms. Recommend resolving the input to an absolute path before building the command (e.g. `"-i", str(video.resolve())`) so the cwd change only affects where the bare-name metadata file is written.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

P2: The new cwd=str(meta_file.parent) on the sampling subprocess changes the directory the ffmpeg -i input is resolved against. video is passed straight through from the caller (grade.py standalone passes the raw, possibly-relative args.input/args.analyze with no .resolve()), so a relative input like python helpers/grade.py input.mp4 -o out.mp4 — a documented usage — will now fail with CalledProcessError because ffmpeg looks for input.mp4 inside the OS temp dir. This also violates the stated "no behavior change on macOS/Linux" goal since it breaks previously-working relative paths on all platforms. Recommend resolving the input to an absolute path before building the command (e.g. "-i", str(video.resolve())) so the cwd change only affects where the bare-name metadata file is written.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At helpers/grade.py, line 127:

<comment>The new `cwd=str(meta_file.parent)` on the sampling subprocess changes the directory the ffmpeg `-i` input is resolved against. `video` is passed straight through from the caller (grade.py standalone passes the raw, possibly-relative `args.input`/`args.analyze` with no `.resolve()`), so a relative input like `python helpers/grade.py input.mp4 -o out.mp4` — a documented usage — will now fail with CalledProcessError because ffmpeg looks for `input.mp4` inside the OS temp dir. This also violates the stated "no behavior change on macOS/Linux" goal since it breaks previously-working relative paths on all platforms. Recommend resolving the input to an absolute path before building the command (e.g. `"-i", str(video.resolve())`) so the cwd change only affects where the bare-name metadata file is written.</comment>

<file context>
@@ -101,15 +109,24 @@ def _sample_frame_stats(
         ]
-        subprocess.run(cmd, check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+        subprocess.run(
+            cmd, check=True, cwd=str(meta_file.parent),
+            stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
+        )
</file context>