CCExtractor / CCExtractor/taskwarrior-flutter
Repository bloated (~100MB+ clone) by committed Rust native build artifacts
- Dominant language
- Dart
- Stars
- 244
- Forks
- 179
- Avg merge
- 12h 42m
- Merged PRs (30d)
- 2
Description
## Problem
A fresh clone of this repository is **100MB+**, and `.git` alone is ~129MB. The
bloat comes almost entirely from **compiled native libraries that are checked
into git**. These are build outputs of the `rust/` crate (`tc_helper`) and are
fully regenerable, so they should never have been committed.
## Where the size comes from
Largest blobs across history (`git rev-list --objects --all | git cat-file --batch-check`):
| Blob | Approx size (per copy) | Copies in history |
| --- | --- | --- |
| `ios/tc_helper.xcframework/.../ios-arm64_x86_64-simulator/.../tc_helper` | ~48 MB | 3 |
| `android/app/src/main/jniLibs/arm64-v8a/libtc_helper.so` | ~29–30 MB | 3 |
| `ios/tc_helper.xcframework/ios-arm64/.../tc_helper` | ~23 MB | 3 |
| `android/app/src/main/jniLibs/armeabi-v7a/libtc_helper.so` | ~21 MB | 2 |
| `android/main/app/src/main/jniLibs/arm64-v8a/libtc_helper.so` (stray path) | ~9 MB | 1 |
| `android/app/src/main/jniLibs/**/libcrc_fast-*.so` | small | a few |
All of these are produced by building the `rust/` crate (`cargo ndk` for
Android, an xcframework build for iOS). They are not source.
## Proposed fix (two parts)
### Part 1 — stop tracking them going forward (PR, mergeable now)
A PR is on the way that:
- removes the tracked binaries from the working tree,
- git-ignores `android/app/src/main/jniLibs/` and `ios/tc_helper.xcframework/`,
- documents and scripts how to regenerate them (`rust/build_android.sh`,
`rust/build_ios.sh`, plus updated `rust/README.md`).
This stops **future** bloat but does **not** shrink existing clone size on its
own, because the old blobs still live in history.
### Part 2 — reclaim existing history (maintainer action, force-push)
To actually shrink the repo, the historical blobs must be stripped, which
**rewrites every commit SHA and requires a force-push to `main`**. This cannot
be done via a normal PR — only a maintainer can do it. Recommended procedure
using [`git-filter-repo`](https://github.com/newren/git-filter-repo):
```bash
# Work on a fresh mirror clone so nothing local is at risk.
git clone --mirror https://github.com/CCExtractor/taskwarrior-flutter.git
cd taskwarrior-flutter.git
# Strip the build-artifact paths from ALL history.
git filter-repo \
--invert-paths \
--path ios/tc_helper.xcframework \
--path android/app/src/main/jniLibs \
--path android/main/app/src/main/jniLibs
# Inspect the result (repo size, history) before pushing.
git count-objects -vH
# Publish the rewritten history (overwrites all refs on the remote).
git push --force --mirror
```
A size-based alternative that catches any other stray binaries:
`git filter-repo --strip-blobs-bigger-than 5M` (verify it doesn't drop
legitimate assets such as fonts or app icons first).
**Caveats for Part 2 — please read before running:**
- Every commit SHA changes. All open PRs must be rebased/recreated, and every
fork diverges and must re-sync (or re-fork).
- Best done **right after** the Part 1 PR is merged, and ideally announced to
contributors in advance.
- GitHub keeps old objects reachable for a while; you may want to ask GitHub
Support to run `gc` afterward to fully reclaim space on the remote.
Happy to adjust the PR or the rewrite paths as needed.
Contributor guide
Assessment
This issue has not been assessed yet.