fix: auto-detect deleted files in --stage-record-file

[frankieyan]

The fix in PR #34 for removing deleted files from the records file relied on lint-staged passing deleted file paths as arguments. However, lint-staged only provides modified files (add/change), not deleted files. This meant deleted files were never actually removed from records.

This PR changes runStageRecords to auto-detect deleted files by scanning all recorded file paths and checking which ones no longer exist on disk.

Fixes #31

Test plan

1. Build the dist files: npm run build
2. cd src/__fixtures__/sample-project
3. Run node ../../../dist/index.mjs --overwrite to create the records file
4. Manually add a fake deleted file entry to .react-compiler-records.json:

   "src/deleted-file.tsx": { "CompileError": 2 }

5. Run node ../../../dist/index.mjs --stage-record-file src/good-component.tsx (without the deleted file)
6. Verify the output shows the deleted file being removed and that it's gone from the records file

[doist-bot]

This PR successfully introduces logic to auto-detect deleted files in records, addressing the lint-staged limitation. However, there is a critical regression in src/index.mts where a variable rename likely breaks the subsequent record-updating code, and an issue in the README.md script that prevents the tool from running on deletion-only commits.

[doist-bot]

Removing D (Deleted) from the diff-filter causes FILES to be empty for commits that only contain file deletions. In such cases, the script exits early (line 136) and the tracker never runs, so the records are not updated. You should keep D in the filter to ensure the tool is triggered; the new logic in runStageRecords handles non-existent input files gracefully.

[frankieyan]

lint-staged by default also doesn't filter by deleted files, so for todoist-web, commits with only deletes also fail to run esplint.

Maybe asking users to always pass in deleted paths would be the better call (e.g. lint-staged --diff-filter=ACMRD), but not all of our tooling are able to handle it:

[nats12]

Run node ../../../dist/index.mjs --overwrite to create the records file

Hey @frankieyan, not sure if I'm doing something wrong here but that command doesn't create any records file for me. So I thought I'll go ahead and manually create react-compiler-records.json inside /src/__fixtures__/sample-project, then add the deleted file entry from the test plan to that, but I get:

🔍 Checking 1 file for React Compiler errors and updating records…
The following paths are ignored by one of your .gitignore files:
src/__fixtures__/sample-project/.react-compiler.rec.json
hint: Use -f if you really want to add them.
hint: Turn this message off by running
hint: "git config advice.addIgnoredFile false"
⚠️  Failed to stage records file at /Users/nataliemclaren/Sites/Doist/react-compiler-tracker/src/__fixtures__/sample-project/.react-compiler.rec.json

[frankieyan]

@nats12 for the first issue where no records file was created, was there an error message? You shouldn't have to create the file manually. The test suite actually creates this file multiple times before deleting it in the afterEach hook. Here's what the output should look like:

For the second issue, my bad I forgot to mention this will have to be temporarily removed from gitignore, because it's a test artifact, and as it could change between tests, I felt we shouldn't include it in the repo. Once it's able to stage the file, you should see something like this:

[nats12]

@nats12 for the first issue where no records file was created, was there an error message? You shouldn't have to create the file manually. The test suite actually creates this file multiple times before deleting it in the afterEach hook. Here's what the output should look like:

@frankieyan I tried this again, there is no error. My output looks like yours above for step 1. See video below:

CleanShot.2026-01-23.at.16.42.25.mp4

[frankieyan]

@nats12 I see the records file in your file tree (and I realize I got the name wrong in the PR description, sorry!):

I think we can improve the UX here by including the file name in the output

[frankieyan]

@nats12 I added the file path to the output whenever it's being written to. Hopefully that makes it clearer!

If you encounter more issues, please let me know, thanks!