fix(series): count split editions of an owned book as covered, not missing (#2524)

[gchahcg]

Summary

Some Hardcover catalogues list a novel's own split-into-parts editions beside the whole novel — Stormlight's The Way of Kings at series position 1, with its "Part 1"/"Part 2" at 1.1/1.2. Already owning the whole novel left those parts reading as missing volumes on the Series page: counted against the series, and sitting behind an add button that would create and queue a download for text already on the shelf. This is the agreed first narrow slice of #2524 — the "split editions of an owned book" shape from the issue thread's real Stormlight/Wheel of Time catalogue data. Refs #2524 (not Closes: the broader work-identity check the issue asks for — reprints, translations, audiobook-slot filling, a reconciliation preview — stays open).

The rule

A catalogue row is reported as covered, not missing, when both hold:

1. Its position is the fractional child of a Present row's integer position (1.1/1.2 under 1).
2. Its title is that Present row's title plus a Part N / Pt. N / Part N of M marker.

Both signals are required on purpose:

Case	Result	Why
"The Way of Kings, Part 1" at 1.1, whole owned at 1	Covered	position + title agree
"Edgedancer" (a real novella) at 2.5, "Words of Radiance" owned at 2	Missing	no Part N marker — position alone is not enough
Wheel of Time "Two Volume Edition": parts at 1.1/1.2, no row at 1 at all	Missing	nothing Present to cover them — they're the real volumes here
"The Great Hunt, Part 2 of 2: New Threads in the Pattern" vs owned "The Great Hunt"	Covered	a subtitle after the marker is ignored

Implementation notes

• internal/seriesmatch: IntegerFloorPosition (is a position fractional, and what's its integer floor) and SplitPartOf (does a title reduce to a whole title plus a part marker) are the two new primitives. HasPartMarker/StripPartMarker moved here from internal/importer/volume.go — one regex, one package, now that series.go needs the same marker.
• internal/api/series.go: buildHardcoverDiff gains a splitEditionWhole check in the unclaimed-catalogue loop, run after the existing box-set prune. A new Covered []seriesHardcoverDiffBook field on the diff response carries the suppressed rows, each pointing at the whole work's LocalBookID/LocalTitle rather than one of its own.
• Both Fill write paths (createMissingHardcoverBooks, createMissingHardcoverBook) iterate diff.Missing only, so Covered rows stop being created and queued automatically — no change needed there.
• web/src/pages/SeriesPage.tsx: covered rows render in their own collapsed <details> section below Missing, each linking to the book that already covers it, no add button.

Where it's mounted

• internal/api/series.go: buildHardcoverDiff, new splitEditionWhole helper right after it.
• internal/seriesmatch/match.go: IntegerFloorPosition, SplitPartOf, HasPartMarker, StripPartMarker.
• internal/importer/volume.go: volumeTitles now calls the moved seriesmatch helpers instead of its own copy.
• web/src/api/series.ts, web/src/pages/SeriesPage.tsx.

Checklist

• Tests added or updated — seriesmatch table tests for both new primitives; a direct unit test of splitEditionWhole against crafted Present rows (covers the Uncertain/Missing-whole case without needing to engineer a fuzzy title score); two buildHardcoverDiff handler tests (the Stormlight shape, the Wheel-of-Time "no unsplit whole" trap); a Fill write-path test; a SeriesPage render test
• Doc-update gate cleared — docs/Hardcover-Series-Wiki.md (What It Does list + a new paragraph after the box-set rule)
• Wiki pages updated if user-facing behaviour changed — same file
• Changelog fragment added — changelog.d/2524-split-editions.md

Test plan

• gofmt -l ., go vet ./..., golangci-lint run (v2.11.4), govulncheck ./...
• go test ./cmd/... ./internal/... (full suite, no race)
• make test-race — all shards pass except internal/db, which hits this sandbox's own slow-hardware timeout unrelated to this change (zero lines of internal/db touched here; reproduces identically in isolation with nothing else running, and the individual test that was mid-flight passes in 3s on its own)
• make smoke
• cd web && npm run lint && npm run typecheck && npm run build && npm test
• /code-review (high effort) on the diff — came back clean

🤖 Generated with Claude Code

https://claude.ai/code/session_01WA2CUNg5o4wUBaX6yWXNNk

[codecov]

Codecov Report

❌ Patch coverage is 97.14286% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
web/src/pages/SeriesPage.tsx	66.66%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[gchahcg]

Closing/reopening to retrigger CI — CodeQL failed on all three languages with Loaded a configuration file for version '4.36.2', but running version '4.38.2', a CodeQL Action version mismatch unrelated to this diff (other branches' CodeQL runs around the same time succeeded).

[gchahcg]

Marking this ready for review. Two red checks are visible but both are unrelated to this diff:

• SAST – Frontend fails on npm audit --audit-level=high, which just started flagging a new high-severity advisory in source-map-js (GHSA-68fv-2mgg-jv7q, event-loop DoS), a transitive dependency. Confirmed this fails identically on main with no changes — a freshly-published supply-chain advisory affecting every open PR right now, not something this PR introduced or touches (no package.json/package-lock.json changes here).
• validate (Go race) (non-api) was cancelled by the job's own 30-minute wrapper mid-run on unrelated internal/importer tests. This diff makes no concurrency changes (it moves two pure string-helper functions). All five other race shards pass; this one is independently non-blocking by the workflow's own fail-fast: false design.

Everything else — full Go suite, lint, govulncheck, smoke, ABS contract, CodeQL (go/js/ts all passed on the latest run after a transient bundle-version mismatch cleared), frontend build/lint/test, licenses, secrets/IaC/container/DAST scans — is green.

[vavallee]

Nice work on this, the two signal rule is the right call and I couldn't make it misfire. Henry IV style titles, numbered volumes and the Two Volume Edition shape all stay missing, and the tests fail on main the way they should. Merged current main into it locally with no conflicts and the api, seriesmatch and importer suites are green.

Two small things before I merge. The changelog fragment has an em dash in it (our README example does too, my fault), so please drop it and trim the entry a bit since it reads as release notes. And the new covered section on SeriesPage has hardcoded English plus a hand rolled plural, while the rest of that page goes through t(). Could you move those strings into en.json under series.hardcover with i18next plurals?

Not for this PR, just so it's written down: anyone who already filled Stormlight before this has the part rows locally, and Fill will keep queueing them. I'll open a follow-up for that.

[gchahcg]

Fixed both:

• Changelog fragment: dropped the em dash and trimmed it down, it was reading like release-note prose rather than a changelog bullet.
• Covered-section strings moved into en.json under series.hardcover (which already existed from Allow inspecting and correcting series positions and catalogue bindings on existing books #2554): coveredSection_one/coveredSection_other (i18next plural, replacing the hand-rolled ? '' : 's') and coveredSubtitle. Extended the test to cover the plural form too.

Also rebased onto current main (#2554's series rewrite landed in the meantime) — no conflicts in the logic, just the two string additions needing to land in the series.hardcover object that already exists now. Full suite green after.

Noted on the Stormlight-already-filled follow-up, that makes sense as its own issue.

[vavallee]

Thanks for both fixes, the en.json plurals and the clean changelog look good. Since this was opened, #3109 landed on main with seriesmatch.SplitEditionPartOf and BookRepo.ListCoveredSplitEditionParts, and its description asks whichever PR lands second to line up with the other.

Could you switch splitEditionWhole over to SplitEditionPartOf and drop IntegerFloorPosition, SplitPartOf and the second copy of the Part N regex? HasPartMarker and StripPartMarker can stay and use the shared regex.

Also, the rule on main only counts a whole that is not excluded and is imported or monitored. Right now this PR covers parts under any present row, so an excluded or unmonitored whole hides its parts here while the sweep still treats them as books. Gating the wholes the same way, with a test for each case, would keep the two in agreement.

Last thing: the wiki now has two paragraphs on this (lines 90 and 98), so merging them into one without the em dashes would be great. Really nice work on the Edgedancer and Wheel of Time cases.

[gchahcg]

Consolidated with #3109:

• splitEditionWhole now calls seriesmatch.SplitEditionPartOf instead of the separate IntegerFloorPosition/SplitPartOf pair, both removed. HasPartMarker/StripPartMarker keep their own direct tests but now share split_edition.go's splitEditionMarkerRe instead of a second copy of the regex.
• Added the eligibility gate: a Present row only counts as a whole covering catalogue-diff parts when it's also not excluded and imported-or-monitored, the same rule ListCoveredSplitEditionParts applies to stored rows. Pulled into a small eligibleSplitEditionWhole helper, cross-referenced in a comment from both sides of the SQL/Go boundary so a future eligibility tweak doesn't update one and miss the other. Added a test for each half of the gate (excluded, wanted-and-unmonitored, wanted-and-monitored).
• Merged the wiki's two split-edition paragraphs into one, no em dashes.

Ran /code-review on the full diff after — the only finding that was actually about this PR's code (as opposed to unrelated pre-existing code the review's diff range happened to sweep in) was this exact duplicated-predicate concern, which the eligibility-gate change above addresses.

Full suite green: gofmt/vet/lint/govulncheck, full Go tests, frontend lint/typecheck/build/tests.

[gchahcg]

The lint and Container Scan failures on the last run are unrelated to this diff: GO-2026-6617 (fixed in go1.26.9) landed in both govulncheck's and Grype's vulnerability databases today, and the repo's pinned go1.26.8 toolchain is affected — this will fail on main's own next CI run too, not just this PR. Opened #3141 to bump the toolchain everywhere make go-version-check tracks (go.mod, all three Dockerfiles, every workflow). Once that merges this PR's CI should go green on its own on the next push/rebase.

[vavallee]

Thanks gch, this is exactly what I was after. Dropping the duplicate helpers and calling SplitEditionPartOf directly, plus the status gate that mirrors the SQL with a comment pointing at it, makes the diff and the sweep agree. The excluded and unmonitored whole tests both fail when I break the gate, which is what I wanted to see.

One small wording thing, non-blocking: "books you own" in the series section title, the wiki and the changelog is a bit off, since a monitored wanted whole counts too. "books in your library" would be closer. Otherwise good to go.

The red lint and Container Scan checks are the Go 1.26.8 issue, fixed on main by #3131, so merging main in will clear them.