ci: set UV_LINK_MODE=copy in the Windows workflow

[mattmillerai]

ELI-5

The Windows CI job installs packages with uv, and uv saves time by hardlinking files instead of copying them. On Windows it cannot overwrite a .pyd that some other process still has open, so the install blows up with "Access is denied" before any test runs. Telling uv to just copy the files makes the problem go away.

What

Adds UV_LINK_MODE: copy to jobs.test.env in .github/workflows/test-windows.yml, alongside the existing PYTHONIOENCODING.

Why

The "Windows Specific Commands" workflow fails on every recent PR across all branches, in the Install Dependencies step, before any test runs:

error: failed to remove file `...\venv\Lib\site-packages\pydantic_core\_pydantic_core.cp312-win_amd64.pyd`: Access is denied. (os error 5)

uv then exits non-zero and the half-removed .pyd produces a downstream ImportError: cannot import name '__version__' from 'pydantic_core'.

comfy install --fast-deps shells out to python -m uv pip install (comfy_cli/uv.py), and uv's default link mode on Windows is hardlink/clone — it cannot replace a .pyd that is loaded/locked by another process on the runner. Copy mode writes a fresh file instead, which is the standard remediation for this error. Setting it at job level (rather than on the step) means it also reaches the uv subprocess that comfy install spawns; comfy_cli/uv.py never passes an explicit env=, so the child inherits it.

Verification / judgment calls

• This PR cannot verify itself. The workflow only triggers on paths: comfy_cli/**, so a workflow-only change does not run it. Verify by re-running "Windows Specific Commands" on any open PR that touches comfy_cli/** (e.g. fix(generate): fetch spec from /openapi not /openapi.yml (BE-2982) #517) once this lands, and confirm Install Dependencies and the subsequent comfy launch --quick-test-for-ci both pass.
• I deliberately did not add .github/workflows/test-windows.yml to the paths filter (which would make the workflow self-testing) — that is a separate behavior change and out of scope here. Flagging it as a reasonable follow-up.
• No Python source is touched, so there is no test or lint surface for this diff; the YAML was parsed to confirm jobs.test.env resolves to {PYTHONIOENCODING: utf8, UV_LINK_MODE: copy}.
• Tradeoff: copy mode is marginally slower/more disk than hardlinking. On a fresh CI runner that is negligible next to a job that currently always fails.

[coderabbitai]

Warning

Review limit reached

You’ve reached a temporary PR review limit under our Fair Usage Limits Policy.

Your recent review volume is higher than typical usage, so adaptive limits are currently applied.

Next review available in: 42 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 9460063a-d6a4-4c63-8b16-a12042e7850f

📥 Commits

Reviewing files that changed from the base of the PR and between a732f5c and 9e43e67.

📒 Files selected for processing (1)

• .github/workflows/test-windows.yml

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch matt/be-3278-uv-link-mode-copy

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch matt/be-3278-uv-link-mode-copy

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

🔍 Cursor Review — Consolidated panel

Triggered by @mattmillerai.

✅ No high-signal findings.

Panel: 6/8 reviewers contributed findings.

Reviewers that did not contribute: kimi-k2.5:adversarial (empty), kimi-k2.5:edge-case (empty)

[mattmillerai]

Reviews-resolver pass: no changes needed on this PR — no unresolved review threads, no merge conflict (`MERGEABLE`), and the diff is still exactly the intended 3 lines in `.github/workflows/test-windows.yml`. Verified `jobs.test.env` resolves to `{PYTHONIOENCODING: utf8, UV_LINK_MODE: copy}`.

The red build check is pre-existing and unrelated to this PR

build fails on tests/comfy_cli/registry/test_config_parser.py (7 tests) and tests/comfy_cli/command/nodes/test_node_init.py::test_node_init_strips_credentials with:

ValueError: Comment cannot contain line breaks
  tomlkit/items.py:504

Root cause: comfy_cli/registry/config_parser.py:75-77 attaches a multi-line comment to [tool.comfy].includes. tomlkit 0.15.0+ rejects line breaks in Item.comment(). CI installs via pip install -e . (unpinned, pyproject.toml:49), so it resolves tomlkit 0.15.1, while uv.lock still pins 0.13.3 — which is why it reproduces in CI but not under uv run.

Evidence this is not caused by this PR:

• With tomlkit==0.13.3 → 81 passed; with tomlkit==0.15.1 → 8 failed, 73 passed.
• Reproduced against main's own code (same 8 failures), so it predates this branch and affects every open PR.
• This PR touches no Python and does not affect build-and-test.yml at all.

This is also real user-facing breakage, not just tests: create_comfynode_config() backs comfy node init, so a fresh install crashes today.

Fixing product code inside a Windows-CI env-var PR would be scope creep, so I've deferred it to a follow-up ticket with a verified fix (comfy.add(tomlkit.comment(...)) preserves the original rendering under 0.15.1) plus a recommendation to make CI installs honor uv.lock. That fix lands on main and unblocks this PR and everything else.

No action needed here beyond a re-run once the tomlkit fix merges.

[mattmillerai]

Correction to my previous comment: the tomlkit fix is already in flight as #533 ("fix(registry): emit pyproject hint blocks as single-line tomlkit comments", BE-3285), so no new follow-up ticket is needed and I have not filed one.

#533 covers the same root cause and is in fact broader than what I sketched — it fixes both multi-line .comment() sites (the requires-comfyui hint and the classifiers block in initialize_project_config), emitting one comment item per line via a shared _append_hint() helper.

So the state of this PR is unchanged: nothing to fix here. Once #533 merges to main, re-run build on this PR and it should go green. Still worth considering separately (not filing unless someone wants it) is that CI's pip install -e . ignores uv.lock, which is why tomlkit 0.15.1 reached CI while local uv run stayed on the locked 0.13.3 — that gap is what let this land unnoticed.

[bigcat88]

Approving, with one correction: the failure this fixes is no longer happening, so this lands as hardening rather than a fix.

The description says the Windows job "fails on every recent PR across all branches." That was true when you opened this (2026-07-17), but the last 12 test-windows.yml runs are 12/12 success, including branches from this morning:

2026-07-30T07:33  success  matt/be-4357-git-utils-cwd
2026-07-30T07:13  success  matt/be-3275-dedupe-auth-header-selection
2026-07-30T06:27  success  matt/be-3266-cloud-http-error-envelope
… (9 more, all success)

The pydantic_core Access is denied (os error 5) breakage was resolved by #572's mixpanel<5 pin (2026-07-22), which stopped the version churn that had uv replacing a loaded .pyd mid-install.

I'm still approving because the underlying condition hasn't gone away — uv's default link mode on Windows remains hardlink/clone, and it still cannot replace a .pyd another process holds open. The pin removed today's trigger, not the mechanism. UV_LINK_MODE: copy costs a little install time on a CI runner and removes that whole failure class. uv's hardlink fallback is live and visible even here on Linux:

warning: Failed to hardlink files; falling back to full copy.
         If this is intentional, set `export UV_LINK_MODE=copy` … to suppress this warning.

Your job-level placement reasoning checks out. I verified the claim rather than assuming: comfy_cli/uv.py has zero env= arguments across its subprocess calls (_check_call is a bare subprocess.check_call(cmd, cwd=cwd)), so the uv child that comfy install --fast-deps spawns inherits the job environment. Putting it on the step instead would have missed exactly that subprocess.

Three lines, CI-only, no runtime code touched. Worth updating the description so the next reader doesn't go looking for a red Windows job that isn't there any more.