Symptom
The tracked working tree is only ~444 KB, but `.git` is ~30 MiB. A first `git clone` (or any push to a new remote with a 10 MB body limit, e.g. a typical Solid pod's git endpoint) transfers all of that history.
Cause
`node_modules/` was committed at some point in the past. It's now in `.gitignore`, but the historical blobs are still in the pack.
Top blobs across all history:
| Size |
Path |
| 11.1 MB |
`node_modules/@esbuild/linux-x64/bin/esbuild` |
| 2.1 MB |
`node_modules/@rollup/rollup-linux-x64-gnu/rollup.linux-x64-gnu.node` |
| 1.3 MB |
`node_modules/vite/dist/node/chunks/config.js` |
| 1.2 MB |
`node_modules/rdflib/dist/rdflib.min.js.map` |
| 930 KB |
`node_modules/rollup/dist/es/shared/node-entry.js` |
| 929 KB |
`node_modules/rollup/dist/shared/rollup.js` |
| 792 KB |
`node_modules/jsonld/dist/jsonld.js` |
| 711 KB |
`node_modules/@asamuzakjp/css-color/dist/browser/css-color.min.js.map` |
| 623 KB |
`node_modules/jsonld/dist/jsonld.esm.js` |
| 521 KB |
`node_modules/rdflib/dist/515.rdflib.min.js.map` |
Repro:
```
git rev-list --objects --all
| git cat-file --batch-check='%(objecttype) %(objectname) %(objectsize) %(rest)'
| awk '$1=="blob"' | sort -k3 -nr | head
```
Impact
- New contributors download ~30 MiB to clone a 444 KB project
- Pushing to git remotes with a body-size limit (~10 MB) fails with HTTP 413 — hit this trying to install the app via `git push` to a Solid pod
- CI fetches and any future `gh-pages` deploys carry the same overhead
Proposed fix
Purge `node_modules/` (and any other historical bloat) from history with `git-filter-repo`:
```
git filter-repo --invert-paths --path node_modules/
```
This rewrites all commit SHAs, so it needs coordination:
- Pick a quiet window
- Run the filter locally on a fresh clone
- `git push --force` to GitHub
- All collaborators `git fetch --all && git reset --hard origin/` (or re-clone)
- Any open PRs need to be rebased
Expected reduction: ~30 MiB → ~1 MiB (444 KB working tree + a small amount of compressed history).
Optional belt-and-braces
After the purge, add a pre-commit / CI guard to prevent regression. Example with husky:
```yaml
.husky/pre-commit
git diff --cached --name-only | grep -q '^node_modules/' && {
echo "refusing to commit anything under node_modules/"; exit 1; }
```
Symptom
The tracked working tree is only ~444 KB, but `.git` is ~30 MiB. A first `git clone` (or any push to a new remote with a 10 MB body limit, e.g. a typical Solid pod's git endpoint) transfers all of that history.
Cause
`node_modules/` was committed at some point in the past. It's now in `.gitignore`, but the historical blobs are still in the pack.
Top blobs across all history:
Repro:
```
git rev-list --objects --all
| git cat-file --batch-check='%(objecttype) %(objectname) %(objectsize) %(rest)'
| awk '$1=="blob"' | sort -k3 -nr | head
```
Impact
Proposed fix
Purge `node_modules/` (and any other historical bloat) from history with `git-filter-repo`:
```
git filter-repo --invert-paths --path node_modules/
```
This rewrites all commit SHAs, so it needs coordination:
Expected reduction: ~30 MiB → ~1 MiB (444 KB working tree + a small amount of compressed history).
Optional belt-and-braces
After the purge, add a pre-commit / CI guard to prevent regression. Example with husky:
```yaml
.husky/pre-commit
git diff --cached --name-only | grep -q '^node_modules/' && {
echo "refusing to commit anything under node_modules/"; exit 1; }
```