Add review-patch command for interactive patch inspection#1290
Open
spoorcc wants to merge 18 commits into
Open
Add review-patch command for interactive patch inspection#1290spoorcc wants to merge 18 commits into
spoorcc wants to merge 18 commits into
Conversation
Introduces `dfetch review-patch` which stages the clean upstream source in the git index and applies the selected patches to the working tree, so any diff-aware editor sees `git diff` (working tree vs index) showing exactly what the patches contribute. The command always restores original state on exit — no permanent changes to working tree or index. - New command: dfetch/commands/review_patch.py with --count/-n, --interactive/-i - GitSuperProject.add_path() and restore_staged() methods for index control - GitLocalRepo.add_path() and restore_staged() as git-level primitives - SVN superprojects supported (with a warning; no staging step) - Interactive TUI uses read_key()/Screen for ← → step-through - 8 unit tests, 4 git BDD scenarios - Documentation and changelog updated Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (2)
WalkthroughAdds a new Changesreview-patch command feature
Auto-generated documentation and security metadata updates
Sequence Diagram(s)sequenceDiagram
actor User
participant CLI as ReviewPatch.__call__
participant SuperRepo as GitSuperProject
participant SubRepo as SubProject
participant TUI as _step_tui
User->>CLI: dfetch review-patch [--count N | -i] [projects]
CLI->>CLI: validate args (TTY check, count >= 0)
CLI->>SuperRepo: create superproject, check VCS type
loop each selected project
CLI->>SubRepo: create_subproject(project)
CLI->>CLI: _can_review_project (patch/fetch/changes checks)
alt skip condition met
CLI-->>User: log warning, skip project
else eligible for review
CLI->>SubRepo: update(patch_count=0)
CLI->>SuperRepo: add_path(project_path)
alt --interactive
CLI->>TUI: _step_tui(patches)
TUI-->>User: _draw_tui_frame + key hints
User-->>TUI: LEFT / RIGHT / ENTER / ESC
TUI-->>CLI: complete or abort
else bulk apply
CLI->>SubRepo: apply_patches(count)
CLI-->>User: "inspect via git diff / svn diff"
end
Note over CLI,SuperRepo: try/finally guaranteed restoration
CLI->>SuperRepo: restore_staged or restore_from_head
CLI->>SubRepo: update(patch_count=original)
CLI-->>User: "Restored original state"
end
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related issues
Possibly related PRs
Suggested labels
Suggested reviewers
🚥 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 docstrings
🧪 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 |
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
- Extract shared project-iteration loop into Command._iter_projects to eliminate duplicate code between review_patch and update_patch (R0801) - Fix import order: review_patch before update in __main__.py (isort) - Remove redundant `import dfetch.project` in review_patch; import create_sub_project directly instead (code review finding) - Remove unused MagicMock import from test_review_patch (code review finding) - Update test patch targets to match new import locations Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 5
🤖 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 `@dfetch/commands/review_patch.py`:
- Around line 113-200: The `_review_project` method has too many branches and
conditional paths, exceeding the cyclomatic complexity limit of 8. Extract the
guard validations (checking for patch existence, on_disk_version, and local
changes) into a separate helper method that returns early if validation fails,
move the interactive review logic and non-interactive logic into their own
helper methods, and simplify the main method to orchestrate validation,
application, and restoration in a clearer sequence. This will distribute the
branching logic across focused helper methods while keeping the main method as a
clear orchestrator.
- Around line 70-78: The --count argument lacks validation, allowing negative
integers that produce unexpected Python slice behavior rather than a meaningful
CLI contract. Add a custom type validator to the add_argument call for --count
to ensure only positive integers are accepted. Additionally, update the logic
around line 173 where the raw count is forwarded to patch_count to validate
against negative values and clamp to valid ranges. Finally, modify the reporting
logic around line 181 to track and report the actual number of patches that were
successfully applied, not the requested count, since the effective count may
differ from what was requested.
- Around line 183-187: The logger.print_info_line call in the review_patch
function currently hardcodes the instruction to use git diff, but this is
inconsistent with non-Git superprojects like SVN that should use their own diff
commands. Make the diff command suggestion in the message VCS-aware by checking
the project's VCS type and conditionally including the appropriate diff command
(git diff for Git projects, svn diff for SVN projects, etc.) in the status
message printed to the user.
In `@dfetch/vcs/git.py`:
- Around line 785-793: Add the `--` separator before the path argument in both
the add_path and restore_staged methods to prevent Git from interpreting
option-style paths as flags. In add_path, insert `"--"` between `"add"` and
`path` in the command list passed to run_on_cmdline. In restore_staged, insert
`"--"` between `"--staged"` and `path` in the command list passed to
run_on_cmdline. This ensures Git treats the path as a positional argument rather
than a potential option flag.
In `@tests/test_review_patch.py`:
- Around line 79-97: Add a new test function to validate that negative count
values are rejected by the ReviewPatch command. Create a test that instantiates
ReviewPatch, mocks the required dependencies (create_super_project,
create_sub_project, in_directory, is_tty) similar to
test_review_count_1_uses_patch_count_1, and then calls cmd(_make_args(count=-1))
while asserting that this raises an appropriate validation error or exception.
This ensures the CLI contract maintains count-based validation and prevents
regression to slice-driven behavior for negative values.
🪄 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: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 1290abec-833f-46ef-9d9d-5ff586b6111b
📒 Files selected for processing (11)
CHANGELOG.rstdfetch/__main__.pydfetch/commands/command.pydfetch/commands/review_patch.pydfetch/commands/update_patch.pydfetch/project/gitsuperproject.pydfetch/vcs/git.pydoc/howto/patching.rstfeatures/review-patch-in-git.featurefeatures/review-patch-in-svn.featuretests/test_review_patch.py
- Add `dfetch review-patch` step after `dfetch update-patch` in both the `run` (all platforms) and `test-cygwin` jobs in run.yml - Add `review-patch-demo.sh` cast generation script (mirrors update-patch-demo.sh; pipes stdin to avoid blocking on input()) - Register the demo in generate-casts.sh after the update-patch line - Add synthetic review-patch.cast (asciicast v2) showing the full flow: manifest inspection, patch preview, staging, and restore - Wire `.. asciinema:: ../asciicasts/review-patch.cast` into the Reviewing a patch section of doc/howto/patching.rst Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
In the default (all-patches) case review-patch was fetching three times: once for clean upstream, once to apply all patches, and once more in the finally block to "restore" — even though the working tree was already fully patched after the second fetch. Track worktree_fully_patched after the non-interactive update call and skip the finally fetch when the working tree is already at patch_count=-1. The --count N and --interactive paths still re-fetch because the final working-tree state differs from fully patched. Update tests and the synthetic demo cast accordingly. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
After the initial update(patch_count=0) puts clean upstream in the working tree there is no reason to re-fetch for any subsequent step. - Add SubProject.apply_patches(count) — applies patch files to the already-fetched working tree without going back to the remote. - Add GitSuperProject.restore_worktree(path) / GitLocalRepo.restore_worktree — runs `git restore <path>` to reset the working tree from the staged index (which holds clean upstream throughout the review session). - Non-interactive: replace the second update() call with apply_patches(). Git all-patches case: 1 fetch total (was 2). Git --count N case: 1 fetch + restore_worktree + apply_patches (was 3). - Interactive TUI: replace per-step update() calls with direct patch apply/reverse (Patch.from_file().apply() / .reverse().apply()). No fetch at all during stepping, for both Git and SVN. - Finally block (Git): restore_worktree + apply_patches instead of a re-fetch. SVN still re-fetches when the worktree is not fully patched (no staged index to restore from). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
Extract _can_review_project, _apply_review, _restore_project, _draw_tui_frame, and _apply_step as module-level helpers so every function stays below the CC=8 limit enforced by CI. Black-format tests to match. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
Black reformatted latex_tabs.py, tm_supply_chain.py, tm_usage.py, and related doc/security files during the review-patch session. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
After the refactor, review-patch does only one fetch (clean upstream) and applies/reverses patches directly without re-fetching. The expected output in the BDD scenarios previously included 3 Fetched lines; update them to reflect the new single-fetch flow. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
…ff cmd - Reject --count < 0 with a RuntimeError early in __call__ - Add test_negative_count_raises to prevent regression - Add -- separator before path in git add, git restore --staged, and git restore to prevent option-style paths being parsed as flags - Show `svn diff` in the status message for SVN superprojects instead of always saying `git diff` - Clamp reported patch count to min(requested, total) so the message reflects patches actually applied rather than the raw CLI argument - Add review-patch to the dfetch_cli threat-model description, noting the transient git index mutation and interruption risk - Reduce _apply_review to 5 args by pre-computing chosen_count and info_msg in _review_project (fixes pylint too-many-arguments) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
Remove the second (now-eliminated) fetch from the cast, regenerate with agg to produce an animated GIF showing the single-fetch flow. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
- Replace sp.__class__ = GitSuperProject/NoVcsSuperProject with Mock(spec=...) so isinstance() checks work without unsafe __class__ mutation that pyright rejects - Add str | None type annotation to on_disk_version parameter so passing None in test_never_fetched_logs_warning_and_skips is valid Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
Replace is_git: bool + assert isinstance(superproject, GitSuperProject) pattern with git_super: GitSuperProject | None so the type checker tracks the narrowed type directly — no assert needed. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
update(patch_count=0) writes patch: '' to the metadata file. The restore path fixes the working tree but apply_patches() has no metadata-writing logic, leaving the metadata file missing the patch list. Fix: snapshot the metadata file bytes before the clean-upstream fetch and write them back in the finally block after restore completes. Add a BDD step "the metadata of ... lists patch ..." and assert it in both Git and SVN review-patch scenarios to pin the invariant. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
Instead of asserting the patch field in dfetch_data.yaml directly, verify from the user's VCS perspective: git status --porcelain SomeProject must report nothing after review-patch exits, covering both the working tree files and the metadata file in one shot. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
patch_ng logs "successfully patched N/M: file" via Python logging when Patch.apply() is called. In the TUI loop this output appears between frames, so Screen.draw()'s cursor-up count is wrong and the help line gets duplicated on each arrow keypress. Suppress the patch_ng logger to CRITICAL for the duration of each LEFT/RIGHT apply/reverse operation inside the TUI loop. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
If Patch.apply() or Patch.reverse().apply() raises RuntimeError during the interactive stepper, the TUI frame was left on screen while the error and 'restored' messages printed below it. Catch RuntimeError in _step_tui, clear the screen, then re-raise so the existing try/finally in _review_project still runs _restore_project (working tree and metadata are always restored), and _iter_projects logs the error under the project name as usual. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
The old restore path did restore_worktree (WT = staged = clean upstream) followed by apply_patches(), which would hit the same failing patch and raise again, leaving the index staged, the metadata corrupted, and the working tree in the wrong state. Replace with restore_from_head (git restore --source=HEAD --staged --worktree -- <path>) which atomically resets both the working tree and the index to the committed HEAD state without re-applying patches. This makes restore unconditionally safe regardless of patch health. Also nest the metadata write_bytes in its own try/finally so the dfetch_data.yaml snapshot is always restored even if _restore_project encounters an error (e.g. SVN path where apply_patches can still fail). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
| import argparse | ||
| import tempfile | ||
| from pathlib import Path | ||
| from unittest.mock import ANY, Mock, call, patch |
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
3 participants
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.
Introduces
dfetch review-patchwhich stages the clean upstream source inthe git index and applies the selected patches to the working tree, so any
diff-aware editor sees
git diff(working tree vs index) showing exactlywhat the patches contribute. The command always restores original state on
exit — no permanent changes to working tree or index.
Co-Authored-By: Claude Sonnet 4.6 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_017zY8BoH65KBX6cz7Pm8aeF
Summary by CodeRabbit
New Features
review-patchcommand to interactively inspect patch contributions without permanent changes--countand step-by-step mode with--interactiveDocumentation