ci: Stop downloading the unused NYC taxi CSV

Conversation

1. maaz7409opened this pull request on 16 Oct 2026description.mdAuthor

   Description

   Running pixi run download-data (also part of pixi run setup-dev) fetches a 432 MB zip from S3 and unpacks a 1.5 GB examples/data/nyc_taxi.csv. Nothing in the repo reads that file any more: since #1462 the examples load the taxi data through hvsampledata, which the same script already downloads (hvs.download("nyc_taxi_remote")), and no test, benchmark or notebook refers to nyc_taxi.csv.

   The same download runs in CI, because holoviz_tasks/pixi-install runs download-data by default in the test and docs jobs (only the build jobs turn it off). For example, every unit, core and benchmark job of 38878999117 logs "Downloading NYC Taxi Data 1 of 1".

   This PR removes the pyct.cmd.fetch_data() call from scripts/download_data.py. examples/datasets.yml stays, because datashader.fetch_data() still falls back to the copy shipped inside the package. That function was deprecated in 0.19.0 (#1462), and its warning says it will be removed in 0.20, so datasets.yml can be deleted in the same change that removes it. Existing local copies of the CSV aren't touched.

   It also adds scripts/** to the code paths-filter in test.yaml. Every test job runs download_data.py while setting up its environment, but without scripts/ in the filter, a PR that only changes the script skips the unit, core, benchmark and example jobs, so CI wouldn't have run this change at all.

   Reproduce: run python scripts/download_data.py with an empty cache (XDG_CACHE_HOME=$(mktemp -d)) and no examples/data/nyc_taxi.csv:

   • main: also downloads the zip and writes the 1,510,609,630-byte csv (28.7 s here).
   • this PR: downloads only the geodatasets (60 MB) and hvsampledata (253 MB) files the notebooks use, and nothing new appears in examples/data/.

   I also ran pre-commit run --all-files (all hooks pass), zizmor on test.yaml (no findings), and the getting-started introduction notebook, which uses the taxi data, without the CSV present (3/3 cells pass).

   AI Disclosure

   Tool & Model: Claude Code + Opus 5.5 (high) Usage: Traced which data the download script fetches and what still uses it, ran the script and the taxi notebook before and after, checked the CI logs.

   • I have tested all AI-generated content in my PR.
   • I take responsibility for all AI-generated content in my PR.

   Checklist

   • Tests are Passing.
2. hoxbrocommented on 19 Oct 2026Member

   Thank you.

   I think you could make the AI output a little less verbose. These changes could have been explained in 1 to 2 sentences.

3. hoxbroreviewed on 19 Oct 2026

   .github/workflows/test.yamlline 68

   @@ -65,6 +65,7 @@ jobs:             code:               - 'datashader/**'               - 'examples/**'+              - 'scripts/**'

   • hoxbrocommented on 19 Oct 2026Member

     I don't want to run test CI when doing a change in scripts as it also have unrelated files. I understand why you did it for this change though.

4. maaz7409commented on 19 Oct 2026Author

   Thank you.

   I think you could make the AI output a little less verbose. These changes could have been explained in 1 to 2 sentences.

   Okay sure, will keep this mind.

5. hoxbromerged this pull request on 19 Oct 2026

Copied from GitHub when this site was built; comments by bots are left out.