-- pg_turbovec v1.7.1 -- Phase W-2 split-write reverted.
--
-- This file is a *reference mirror*. The authoritative install
-- script is generated by `cargo pgrx schema`.
--
-- v1.7.1 ships zero SQL surface changes. It reverts the v1.7.0
-- (Phase W-2) `ambuild` reordering after validation on `meh` at
-- 10 M x 1536-d showed the split-write design did not actually
-- lower peak RSS:
--
--   pre-W (v1.5.x):   121 GiB peak RSS, 60 GiB swap, 5048 s
--   post-W (v1.6.0):  22.5 GiB peak RSS, 0 swap, 5052 s
--   v1.7.0 (Phase W-2): 23.04 GiB peak RSS, 2.67 GiB swap, 7748 s
--   v1.7.1 (revert):    behaves as v1.6.0 (build path identical).
--
-- The hypothesis behind Phase W-2 was that dropping the
-- ~7.7 GiB row-major `packed_codes` Vec mid-finalise (via the
-- new `IdMapIndex::take_packed_codes()`) would shave the peak
-- RSS by ~7.7 GiB. It didn't, because `ps -o rss` includes
-- shared_buffers pages that the backend has pinned, and
-- `write_packed_phase` pins the codes pages in shared memory
-- before `take_packed_codes()` runs. The "freed" heap pages
-- just migrated to pinned shared mem; same RSS budget, plus
-- a 53% build-time penalty from the extra GenericXLog flush
-- phase and 2.7 GiB of swap from second-phase allocations
-- colliding with the OS page cache.
--
-- See `benches/results/phase_w_2_validate_meh_10m_2026_05_27.json`
-- and  for the full analysis.
--
-- The wire format does NOT change. The on-disk index format is
-- byte-identical across v1.6.0, v1.7.0, and v1.7.1 (all
-- `MetaPageData::version = 3`). There is no DDL or on-disk
-- migration required, and no REINDEX is needed across any of
-- these versions.
--
-- See docs/UPGRADING.md for the v1.7.0 -> v1.7.1 patch row
-- (no REINDEX needed). This migration file exists only so the
-- upgrade path `ALTER EXTENSION pg_turbovec UPDATE TO '1.7.1';`
-- resolves without an error. It is intentionally empty.