[ecliseice]

Problem

CWA stores a standard Calibre rating and numeric custom columns, but the cover catalog and the Books List cannot order the whole library by those values. A rating can be used to group books, and the table headers for int and float custom columns are not sortable.

Solution

Add one server-side sort for the Calibre rating and for every custom column of datatype int or float. The order is applied to the existing paged queries, before pagination, as a correlated scalar subquery. That keeps the current joins from duplicating rows. Column ids are checked against the library metadata and are never interpolated into SQL.

Features

• Ascending and descending order. In the Books List the first click sorts high to low.
• The catalog dropdown is built from the library: the Calibre rating, plus each numeric custom column under its own name. A column such as #goodreads_rating appears automatically. A column of another datatype does not.
• The same menu is on the cover catalog, search results, and the other book lists that already share that sort toolbar (author, series, publisher, language, category, format, ratings, read/unread, downloaded, archived, top rated, magic shelves).
• Books List: a read-only Rating column, and sortable headers for int and float custom columns.
• Books with no rating or no custom-column value stay last in both directions. A stored Calibre rating of 0 means unrated. 0 in a custom column is a real value.
• Title sort and book id keep equal values in a stable order across pages.
• The chosen catalog order is stored with the other view settings. An unknown or non-numeric key falls back to the default order and is not kept as the stored key.
• Hot Books, Discover, and ordinary shelves are unchanged. Shelf sorting still rewrites the manual shelf order, so it is left alone.
• No metadata database migration and no change to stored ratings.

Testing

• tests/unit/test_rating_sort.py (73 tests): both directions for the rating and for int/float columns, missing values last, 0 in a custom column kept in order, an empty column keeping title order, several columns, a library with no custom columns, paging of the whole library, a filter plus paging, the series join used by the book lists, stable order when only the first sort expression is applied, ignored and read-only columns, invalid keys including SQL fragments, and list_books for ratings and custom_column_<id>
• pytest -m "smoke or unit" locally: 347 passed, 24 skipped
• The same fast suite on the fork: Test Suite run
• Page timing on a 10,000-book library stays in the same range as the existing sorts (about 11–24 ms per page)
• A temporary container on a copy of a real library: cover catalog, search, author list, and Books List (page 1, page 2, and search) for the rating and a numeric column
• Reviewer: confirm the dropdown matches the sort toolbar in the current theme