@pothos/plugin-drizzle@0.18.0

Minor Changes

• 4105614: Tag each value in a compound cursor, so cursors covering more than one column round trip Date and bigint values.

  Compound cursors were serialized with plain JSON, which returns a Date as a string and cannot serialize a bigint at all. Single-value cursors did not have this problem, so it only surfaced on orderings that named more than one column. Cursors written before this change are still read.

  Also fixes two crashes reachable from the same orderings: an ordering value of null now compares with IS NULL rather than throwing out of drizzle, and composite primary keys are looked up against the table's own columns, which the postgres dialect does not use when reporting them.

  Composite primary keys are now used as the tie breaker whether or not their columns are marked notNull(), since SQL makes primary key columns non-nullable regardless. Previously a key declared with primaryKey({ columns: [...] }) alone was skipped, leaving those connections with the non-unique ordering the tie breaker exists to fix.

• 99cf79e: Connection orderBy can now name an extra selected by the same query, rather than only a column:

  query({
    extras: { titleLength: (table) => sql`length(${table.title})` },
    orderBy: { titleLength: "asc" },
  });

  Pothos orders by the expression, builds the cursor from the value it returns, and compares the expression when paging.

  This gives cursors a way to carry a value the column's JavaScript mapping does not preserve. A timestamp({ mode: 'date' }) column is returned as a Date, which only holds milliseconds, so a cursor built from it cannot address a row stored with microsecond precision. Selecting the full value as an extra and ordering by that can. See "Ordering and cursors" in the drizzle plugin docs.

Patch Changes

• d5211ea: Add the primary key to connection orderBy when the ordering is not already unique.

  A cursor records a row's position in an ordering. When that ordering is not unique, orderBy: { createdAt: 'desc' } for example, rows sharing a value can be returned in a different order on each query, so paging through the connection may return a row twice, or skip it.

  Connections ordered by a unique set of non-nullable columns are unchanged. Other connections get the primary key as a trailing order column, and their cursors include the matching values. Cursors created before this change still work: they page from the columns they contain, and the cursors returned by that page contain the full set.