docs(M01): M01_run2 analysis - paths assert, cache seeding required

Made-with: Cursor
This commit is contained in:
Michael Cahill 2026-03-07 16:03:31 -08:00
parent 696371932a
commit 1bbc0f4e59

View file

@ -0,0 +1,66 @@
# M01 CI Run 2 — Analysis
**Date:** 2025-03-07
**Branch:** m01-ci-truthfulness
**Trigger:** Skip prepare-environment (31588a1e)
---
## 1. Workflow Identity
| Workflow | Run ID | Status |
|----------|--------|--------|
| Linter | 22809961294 | ✓ success |
| Tests | 22809961290 | ✗ failure |
---
## 2. Result
**Setup environment** and **Smoke startup** passed (--exit handling works).
**Start test server** step completed (process launched in background).
**Run tests** failed: wait-for-it timed out (server never bound to 7860).
---
## 3. Root Cause
From output.txt artifact:
```
ModuleNotFoundError: No module named 'torch'
```
With `--skip-prepare-environment`, we never run `prepare_environment()`, so torch/clip/requirements were never installed.
---
## 4. Fix Applied
Added **Install runtime dependencies** step (commit 69637193):
- pip install torch, torchvision (CPU)
- pip install CLIP (--no-build-isolation)
- pip install open_clip
- pip install -r requirements_versions.txt
---
## 5. Remaining Blocker
`paths.py` asserts `sd_path` exists (repositories/stable-diffusion-stability-ai). The app requires the `repositories/` folder. With `--skip-prepare-environment`, we never clone. Repositories must come from **cache**. Cache key `repos-2023-12-30` has never been populated (clone has not succeeded).
**Options to seed cache:**
- workflow_dispatch job with PAT for cross-org clone
- One-time local bootstrap + cache upload (if supported)
- Add repositories as submodules
---
## 6. Run 3 (69637193): Confirmed
**Error:** `AssertionError: Couldn't find Stable Diffusion in any of: [repositories/stable-diffusion-stability-ai, ...]`
Torch/clip/requirements now install correctly. The app requires `repositories/` (paths.py). Cache `repos-2023-12-30` is empty. Must be seeded for CI to pass.