V185: currency-freeze columns on carts, cart items and orders (stacked on #785)

[timujinne]

V185: currency-freeze columns on carts, cart items and orders

Stacked on #785 (V184). Companions: BeamLabEU/phoenix_kit_billing#32 and the ecommerce stage-Э1 PR, which read these columns through their schemas (Order, Cart, CartItem).

What V185 does

• phoenix_kit_shop_carts: base_currency varchar(3), exchange_rate numeric(15,6); phoenix_kit_shop_cart_items: base_unit_price numeric(15,2); phoenix_kit_orders: base_currency varchar(3), exchange_rate numeric(15,6), base_total numeric(15,2). All nullable, no default — ADD COLUMN IF NOT EXISTS, no table rewrite.
• A DO block raises unless phoenix_kit_currencies has exactly one is_default row (the billing chain's V2 partial unique index guarantees it; the guard makes the failure visible before any backfill).
• Backfill without literals: base_currency from the is_default row; exchange_rate = 1.0 when the record's currency is the base, else that currency's table rate (NULL when unknown — honest "unknown rate"); base_unit_price/base_total only for records already in the base currency, NULL otherwise (a conversion nobody performed is not fabricated).
• down/1 drops the six columns and stamps 184 — irreversible in meaning (orders forget the rate they were priced at), said in the comment.
• expected_schema.ex: the six columns are declared by hand (V171 "DECLARED POST-GENERATION" precedent; positions/types from the live ordinal_position), then chain_hash restamped — the full generator aborts on this environment (see V184: remove the dead shop_currency setting #785 for the sync_shop_*_slugs mode-shape mismatch).

Tests

test/phoenix_kit/migrations/v185_test.exs (backfill from the default row; refusal with two defaults). Migrations dir + release_check tests: 16 doctests, 450 tests, 0 failures; expected_schema/ tests 67/0; mix phoenix_kit.release_check Migration Version Sync passes (V135..V185 contiguous, chain hash over 51 files). Applied on a live host: 94/94 carts backfilled USD/1.000000, 11/11 items base_unit_price = unit_price, the one order base_total = total, marker 185. Version bump and CHANGELOG left to the maintainer.

[timujinne]

Independent review (read-only reviewer, separate from the author), range 2fe7f3a5..6668c93f. Verdict: PASS, no findings against V185.

Verified directly

• Six ADD COLUMN IF NOT EXISTS statements match the spec (types, nullability, no defaults); the three backfills derive every value from the is_default row (1.0 for base-currency records, the table rate otherwise, NULL when unknown; base_unit_price/base_total only for base-currency records), all gated WHERE … IS NULL for idempotency; the DO block raises before any column is touched unless exactly one default exists; down/1 drops the six and stamps 184.
• Manifest entries hand-declared under the existing "DECLARED POST-GENERATION" lane (V165/V168/V171/V175 precedent); positions and types cross-checked against the live information_schema.columns — exact match; backfill: nil is the right analogue of V183's target_uuid (no column default). chain_hash restamped (7aef0122…), release_check Migration Version Sync passes (V135..V185 contiguous, 51 files).
• Tests rerun: migrations dir + release_check tests 16 doctests / 450 tests / 0 failures; expected_schema/ 67/0; format --check-formatted and credo --strict clean. The "two defaults" test drops billing's partial index inside its sandboxed transaction so V185's own guard is what fires — legitimate.
• Live host after phoenix_kit_update_v184_to_v185: 94/94 carts USD/1.000000, 11/11 items base_unit_price = unit_price, the one order base_total = total, marker 185, all domains 200.
• Commit by the author only; no version/CHANGELOG changes.

Note, not a blocker: the merge that carried this onto the host's stand branch also pulled four unrelated upstream commits (2.15.0 release, B2B registration setting, admin naming); the two behaviour changes 2.15.0 flags (user_dashboard_enabled default, admin_path wording) do not affect that host.

[ddon]

@timujinne and another conflict is here

[ddon]

@timujinne still conflict here