Skip to content

Commit 0cf4fa5

Browse files
andreas16700claude
andreas16700
and
claude
committed
Run 58: Match upstream 4f9f32a (PR matplotlib#30808) - style-iterator
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
1 parent 4f9f32a commit 0cf4fa5

17 files changed

Lines changed: 4536 additions & 349 deletions

‎.github/workflows/tests.yml‎

Lines changed: 193 additions & 349 deletions
Large diffs are not rendered by default.

‎CLAUDE.md‎

Lines changed: 156 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,156 @@
1+
# Claude Code Instructions for matplotlib ezmon Data Collection
2+
3+
## Quick Start
4+
5+
**Read the full process document first:**
6+
```
7+
Read /tmp/matplotlib/DATA_COLLECTION_PROCESS.md
8+
```
9+
10+
## What This Is
11+
12+
This is a **matplotlib fork** used to validate **ezmon** (our pytest-testmon fork with NetDB support). We replay upstream CI history and verify ezmon correctly selects tests.
13+
14+
## Current Task
15+
16+
Continue processing workflow runs from the tracking table in `DATA_COLLECTION_PROCESS.md`.
17+
18+
## Key Commands
19+
20+
```bash
21+
# Check current state
22+
git log --oneline -3
23+
gh run list --workflow=tests.yml --limit 3
24+
25+
# See remaining work
26+
git fetch upstream
27+
git log --oneline 9b61b471d0..upstream/main --first-parent | wc -l
28+
```
29+
30+
## Important Files
31+
32+
| File/Directory | Purpose |
33+
|----------------|---------|
34+
| `DATA_COLLECTION_PROCESS.md` | **Main documentation** - process, tracking table |
35+
| `reports/` | **Detailed run reports** - organized in files of 5 runs each |
36+
| `.github/workflows/tests.yml` | Our ezmon-enabled workflow (preserve this!) |
37+
| `scripts/` | Automation scripts |
38+
39+
## Report File Structure
40+
41+
Reports are stored in `reports/` directory, organized by groups of 5:
42+
- `reports/runs_001-005.md` - Runs 1-5 (includes historical Commit 1)
43+
- `reports/runs_006-010.md` - Runs 6-10
44+
- `reports/runs_011-015.md` - Runs 11-15
45+
- `reports/runs_016-020.md` - Runs 16-20
46+
- (continue pattern for future runs)
47+
48+
## Critical: Reset Workflow
49+
50+
When processing each run, you must reset to upstream state. **ALWAYS save our files first:**
51+
52+
```bash
53+
# 1. SAVE before reset (including reports directory!)
54+
cp .github/workflows/tests.yml /tmp/our-tests.yml
55+
cp DATA_COLLECTION_PROCESS.md /tmp/DATA_COLLECTION_PROCESS.md
56+
cp CLAUDE.md /tmp/CLAUDE.md
57+
cp -r reports /tmp/our-reports
58+
cp -r scripts /tmp/our-scripts 2>/dev/null || true
59+
60+
# 2. Reset (this wipes our files!)
61+
git reset --hard $UPSTREAM_SHA
62+
63+
# 3. RESTORE after reset
64+
cp /tmp/our-tests.yml .github/workflows/tests.yml
65+
cp /tmp/DATA_COLLECTION_PROCESS.md DATA_COLLECTION_PROCESS.md
66+
cp /tmp/CLAUDE.md CLAUDE.md
67+
mkdir -p reports && cp -r /tmp/our-reports/* reports/
68+
mkdir -p scripts && cp -r /tmp/our-scripts/* scripts/ 2>/dev/null || true
69+
70+
# 4. Commit and push
71+
git add .github/workflows/tests.yml DATA_COLLECTION_PROCESS.md CLAUDE.md reports/ scripts/
72+
git commit -m "Run N: Match upstream $UPSTREAM_SHA (PR #XXXX) - description"
73+
git push origin main --force
74+
```
75+
76+
## Required Git Diffs in Reports
77+
78+
For EVERY run report, include TWO git diffs:
79+
80+
### 1. Upstream Code Changes Diff
81+
Shows what changed between the previous run and current run:
82+
```bash
83+
git diff $PREVIOUS_UPSTREAM_SHA $CURRENT_UPSTREAM_SHA -- '*.py'
84+
```
85+
86+
### 2. Code Parity Verification Diff
87+
Shows that our commit matches upstream (only infrastructure files should differ):
88+
```bash
89+
git diff $UPSTREAM_SHA $OUR_COMMIT --stat
90+
git diff $UPSTREAM_SHA $OUR_COMMIT --name-only -- '*.py' '*.pyi' # Should be empty!
91+
```
92+
93+
Include both diffs in the report under:
94+
- **Git Diff (upstream code changes: Run N-1 → Run N)**
95+
- **Git Diff (code parity: our commit vs upstream)**
96+
97+
## Code Parity Verification
98+
99+
**CRITICAL**: Before pushing, ALWAYS verify:
100+
```bash
101+
# These must return 0 / empty:
102+
git diff $UPSTREAM_SHA HEAD -- '*.py' '*.pyi' | wc -l # Must be 0
103+
git diff $UPSTREAM_SHA HEAD -- lib/matplotlib/ | wc -l # Must be 0
104+
105+
# Only these files should differ:
106+
git diff $UPSTREAM_SHA HEAD --name-only
107+
# Expected: .github/workflows/tests.yml, CLAUDE.md, DATA_COLLECTION_PROCESS.md, reports/*, scripts/*
108+
```
109+
110+
## Commit Efficiency
111+
112+
**Bundle documentation updates with run commits** to avoid redundant CI runs:
113+
- Do NOT commit report updates separately
114+
- Include report updates in the NEXT run's commit
115+
- This prevents triggering extra workflow runs for doc-only changes
116+
117+
## Comparing Results with Upstream
118+
119+
**Our test matrix (5 variants):**
120+
- macOS: Python 3.11/3.12 on macos-14, Python 3.13 on macos-15
121+
- Linux: Python 3.12 on ubuntu-22.04 (added Run 56)
122+
- Linux ARM: Python 3.12 on ubuntu-24.04-arm (added Run 56-retry)
123+
124+
**Compare matching variants** when checking upstream results:
125+
- macOS variants: `gh api repos/matplotlib/matplotlib/actions/runs/ID/jobs --jq '.jobs[] | select(.name | test("macos")) | {name, conclusion}'`
126+
- Linux variants: `gh api repos/matplotlib/matplotlib/actions/runs/ID/jobs --jq '.jobs[] | select(.name | test("ubuntu")) | {name, conclusion}'`
127+
128+
**Note**: Run 56 is the baseline for ubuntu-22.04. Run 56-retry adds ubuntu-24.04-arm baseline.
129+
130+
## Investigation Requirements
131+
132+
For EVERY run, investigate thoroughly:
133+
1. **Check upstream job results** - compare with our matching platform variants
134+
2. **Investigate any discrepancies** - different test counts, failures, etc.
135+
3. **Check for NetDB race conditions** - compare changed file counts across parallel jobs
136+
4. **Document findings** - include investigation results in reports
137+
138+
## Do NOT
139+
140+
- Modify code in `lib/matplotlib/` beyond what upstream has
141+
- Run `git reset --hard` without saving our files first (including reports/)
142+
- Push without preserving our workflow file, docs, AND reports
143+
- Skip the code parity verification
144+
- Skip including both git diffs in reports
145+
- Forget to save/restore the `reports/` directory during reset
146+
- Commit report updates separately (bundle with next run)
147+
- Compare with non-matching upstream variants (use macOS jobs only)
148+
149+
## Session Initiation
150+
151+
Future sessions should be started with:
152+
```bash
153+
cd /tmp/matplotlib && claude
154+
```
155+
156+
Then say: "Continue the matplotlib ezmon data collection process."

0 commit comments

Comments
 (0)