[vavallee]

Closes #2819

Summary

BookRepo.SetFilePath fell back to a bare UPDATE books SET file_path, status on any GetByID failure. That dropped the load error, marked the book imported with no book_files row and no refreshBookStatus, and returned nil for a missing book (0 rows matched).

History: the fallback arrived with dual format support (0e0589c, #23) as a guard around the old single column write. book_files (3322c58, #343) made it obsolete. For a missing row it was a no op that claimed success; for an unreadable row it wrote state the rest of the system no longer reads from. There is no genuine case left, so it's removed:

• a GetByID error is returned, wrapped as load book N: ...
• a missing book returns book N not found

The one production caller, handleNewWantedBook, already logs a returned error as authors: record existing file path with book_id, so these now show up in the log instead of vanishing. Its behaviour otherwise is unchanged (it still returns fileFound=true and skips the auto search).

Overlap with #2794: that PR edits this same fallback (adds updated_at, switches to r.exec) and adds TestBookRepoUpdateHydratedMetadataLegacyUpdatedAtShapes/*/SetFilePathFallback, which expects SetFilePath to succeed on a corrupt row. Whichever lands second needs to drop that hunk and that subcase; with this change there is no fallback write for hydration to observe.

How it was verified

New internal/db/books_setfilepath_test.go uses the technique from the issue (calibre_id='corrupt', plus a missing id). On origin/main:

=== RUN   TestSetFilePathReportsLoadErrors/unreadable_book
    books_setfilepath_test.go:43: SetFilePath returned nil although the book could not be loaded
    books_setfilepath_test.go:51: row was written behind book_files: status="imported" file_path="/books/unreadable.epub", want "wanted" and empty
=== RUN   TestSetFilePathReportsLoadErrors/missing_book
    books_setfilepath_test.go:59: SetFilePath returned nil for a book that does not exist
--- FAIL: TestSetFilePathReportsLoadErrors (0.09s)

Passes after the fix. Mutation checks (both mutants compile):

• if err != nil changed to if false && err != nil: unreadable_book fails, the error becomes book 1 not found and no longer carries the scan error
• return fmt.Errorf("book %d not found", id) changed to return nil: missing_book fails

go build ./... && go vet ./... clean; go test ./cmd/... ./internal/... exit 0.

🤖 Generated with Claude Code

https://claude.ai/code/session_016fJcCVbNnKsmj2MAAMwWj9

[github-actions]

The fix is correct and the test coverage is solid.

Logic
GetByID returns (nil, nil) for a missing row (books.go:461-462) and (nil, err) for a scan failure — the two separate guards now handle both correctly. The old collapsed check fell back to a write that was a no-op for missing rows (reported success on 0 rows matched) and silently bypassed book_files + refreshBookStatus on scan errors. Both cases are now surfaced.

Tests
internal/db/books_setfilepath_test.go covers both failure modes, verifies the wrapped error carries the original scan error, and additionally asserts no dirty state was written to books.status / file_path. The precondition check on GetByID at line 40-43 is a good guard — the test fails loud if the corrupt-row behaviour ever changes.

One flag: conflict with #2794
The PR body calls this out, but worth repeating: #2794 touches the same fallback block (adds updated_at, switches to r.exec) and adds a subcase SetFilePathFallback under TestBookRepoUpdateHydratedMetadataLegacyUpdatedAtShapes that expects the fallback to succeed on a corrupt row. If #2794 lands after this one, that subcase will fail and the hunk needs to be dropped. No blocker — just needs a rebase on whichever merges second.

Otherwise clean.

— 🤖 Bindery triage bot (automated). Reply to correct me; a human will see it.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!