Bump reflexio-ai to 0.2.25 (CPU-default reranker) + venv-direct backend with vendored-install self-repair#72
Merged
Merged
Conversation
… Reflexio Replace editable vendored Reflexio installs with --reinstall --no-deps so the venv always matches the shipped vendor snapshot, invoke the prepared venv's interpreter directly from hooks/CLI/backend instead of uv run, resolve plugin-root to a physical path, and add ensure_vendored_reflexio_active to repair a stale reflexio-ai install before backend start.
reflexio-ai 0.2.25 defaults the relevance-floor cross-encoder to CPU (REFLEXIO_RERANK_DEVICE override available), eliminating ~3 GB of MPS memory the backend accumulated on Apple Silicon. Document the new env var in DEVELOPER.md and return the lock to the strict PyPI source.
|
Important Review skippedReview was skipped due to path filters ⛔ Files ignored due to path filters (1)
CodeRabbit blocks several paths by default. You can override this behavior by explicitly including those paths in the path filters. For example, including ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 WalkthroughWalkthroughUpgrade reflexio-ai to 0.2.25, replace editable vendored installs with non-editable reinstalls, add a vendored-version check/repair at backend start, run Reflexio/CLI/hooks via the prepared venv Python with PYTHONPATH wiring, canonicalize plugin-root symlink targets, and update tests. ChangesReflexio Installation and Execution Refactoring
Sequence Diagram(s)sequenceDiagram
participant BackendScript as backend-service.sh
participant PreparedVenvPython as prepared venv Python
participant UVpip as uv pip
participant ReflexioCLI as reflexio.cli
BackendScript->>PreparedVenvPython: ensure_vendored_reflexio_active() (read vendor pyproject.toml)
PreparedVenvPython->>BackendScript: compare installed reflexio-ai version
BackendScript->>UVpip: uv pip install --reinstall --no-deps vendor/reflexio (on mismatch)
BackendScript->>PreparedVenvPython: set PYTHONPATH and exec -m reflexio.cli services start
PreparedVenvPython->>ReflexioCLI: start backend service
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@plugin/scripts/backend-service.sh`:
- Around line 186-213: The function ensure_vendored_reflexio_active currently
hardcodes plugin_python to "$PLUGIN_ROOT/.venv/bin/python" which breaks Windows;
replace that assignment to use the existing helper claude_smart_plugin_python
(same name used in cli.sh) to resolve the platform-specific venv python path,
keep the rest of the function unchanged (references:
ensure_vendored_reflexio_active, plugin_python, claude_smart_plugin_python) so
the subsequent Python check and pip install use the correct executable on
Windows and Unix.
- Around line 315-320: The call to claude_smart_spawn_detached hardcodes
"$PLUGIN_ROOT/.venv/bin/reflexio" which breaks Windows; compute a platform-aware
reflexio binary path (like plugin_python is resolved) into a variable (e.g.,
reflexio_bin) that picks ".venv/bin/reflexio" on Unix and
".venv/Scripts/reflexio" or ".venv/Scripts/reflexio.exe" on Windows, then
replace the hardcoded path in the claude_smart_spawn_detached invocation with
that reflexio_bin variable so the script works cross-platform.
In `@plugin/scripts/hook_entry.sh`:
- Around line 6-7: Update the header comment in hook_entry.sh to correctly state
that vendored Reflexio is installed non-editably (using the --reinstall
--no-deps style) rather than as an editable install; reference the
smart-install.sh behavior and the relevant tests
(test_reflexio_vendor_release_uses_generated_bundle and
test_smart_install_installs_vendored_reflexio) in the comment so it accurately
describes that hooks use the dependency set produced by smart-install.sh which
includes non-editable vendored Reflexio installs.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 2e2d6475-7631-4097-a393-80e3ffa8f3f6
📒 Files selected for processing (11)
DEVELOPER.mdbin/claude-smart.jsplugin/pyproject.tomlplugin/scripts/backend-service.shplugin/scripts/cli.shplugin/scripts/ensure-plugin-root.shplugin/scripts/hook_entry.shplugin/scripts/smart-install.shreflexio.lock.jsonscripts/release-with-reflexio.shtests/test_install_scripts.py
ensure_vendored_reflexio_active hardcoded the Unix venv python path; resolve it via claude_smart_plugin_python like cli.sh and hook_entry.sh. Launch the backend with the venv python via -m reflexio.cli instead of the Unix-only .venv/bin/reflexio entry point, and build PYTHONPATH with a ;-separator and cygpath-converted vendor path on Windows, since native Windows Python does not understand :-separated MSYS paths.
…-editably smart-install.sh switched the vendored install to --reinstall --no-deps; the header still said hooks rely on editable vendored installs.
The release flow's 'uv lock --project plugin' ran inside the enterprise repo checkout, where plugin/ is a uv workspace member, so it updated the workspace lockfile instead of plugin/uv.lock. Regenerated standalone so 'uv sync --locked' passes in CI.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Two related changes that together fix the multi-GB memory footprint of the claude-smart backend on Apple Silicon and harden how the plugin runs its Python environment.
1. reflexio-ai 0.2.25 (ReflexioAI/reflexio#143)
The relevance-floor cross-encoder previously auto-selected MPS on Apple Silicon; torch's MPS caching allocator accumulated GPU buffers across searches and never released them — measured at 4.4 GB physical footprint (3.0 GB IOAccelerator/MPS) on a live backend. 0.2.25 defaults the reranker to CPU with a
REFLEXIO_RERANK_DEVICEoverride (documented in DEVELOPER.md) and silences tqdm batch bars in server logs.Verified live: after a 12-query /api/search burst on the fixed backend, vmmap shows no IOAccelerator region and a 982 MB footprint (peak 1.2 GB) — a ~3.4 GB reduction — with relevance-floor filtering still active.
Lock returned to the strict PyPI source via
REFLEXIO_RELEASE_SOURCE=pypi scripts/release-with-reflexio.sh(release checks + full test suite + npm pack passed).2. Backend runs from the prepared venv with vendored-install self-repair
--reinstall --no-deps(not editable) so the venv always matches the shipped vendor snapshot.uv run.ensure_vendored_reflexio_activerepairs a stale reflexio-ai install before backend start (covered in tests/test_install_scripts.py).ensure-plugin-root.shresolves the symlink target to a physical path.Tests
Summary by CodeRabbit
Documentation
Chores
Tests