refactor: single-source the posting weight arithmetic in booking - #1736

What

Single-sources the Beancount posting-weight arithmetic. The generic ladder in rustledger-booking (shared by the residual and interpolation since #1700) is now exposed as two public Decimal functions, and the two copies of the rule — the BQL weight column and the currency_accounts plugin — delegate to it.

Why

Phase-1 correctness sweep, residual/weight family. The weight rule existed three times, and the copies had already drifted from the canonical:

• Query (compute_posting_weight): recomputed total-cost weights as units × per_unit via CostSpec::resolve, reintroducing the division-then-multiplication precision loss the booking ladder deliberately avoids by using the preserved total (fix(query): DisplayContext::format pads up but never quantizes down (#1103) #1106/fix(query): split format vs format_amount semantics — recover BQL compat (refs #1112) #1113). Observable: SELECT weight for a booked {{100.00 USD}} over 3 units showed 99.99999999999999999999999999 USD while rledger check balances the same posting at 100.00 exactly — the weight column disagreed with the balance checker it's supposed to explain.
• Plugin (weight_of in currency_accounts): took abs() of totals and @@ prices before re-signing by units, silently "fixing" negative magnitudes instead of following the canonical total × signum.

Same drift category as #1648/#1726: consumers re-deriving a rule the pipeline already owns.

How

New public surface in rustledger-booking, each a thin Decimal instantiation of the existing generic (WeightNum) implementation — no second copy of the arithmetic exists:

• cost_number_weight(units, &CostNumber) — the four-variant cost arithmetic (PerUnit multiply, Total/PerUnitFromTotal preserved total × signum, Compound N·a+b per booking: per-unit-from-total cost ({a # b}) weight is wrong in all three forms #1700).
• price_weight(units, price, PriceKind) — @ = |units| × price × signum, @@ = price × signum (the BQL compat: weight() sign flip on total_annotation.beancount (1 mismatch) #1052 credit-side flip; zero units weigh zero).

residual_weight's inline price arm now goes through the same generic helper. The query keeps its fallback order (cost → price → units, Null without units) and drops its now-unused txn_date parameter; the plugin keeps its DTO parsing, with the exhaustive compile-fail match preserved as the CostNumberData → CostNumber mapping.

Edge-behavior deltas in the copies (all toward the canonical): query's zero-units @@ weight is now 0 instead of +total; plugin's negative-magnitude totals are no longer abs'd (the validator rejects negative cost totals upstream, so this is unreachable in the pipeline).

Deep-review addenda (post-publication):

• A fourth zero-units delta: booking skips the Total → PerUnitFromTotal rewrite for zero-unit postings, so 0 X {{100 USD}} reaches the executor as raw Total. Old code: CostSpec::resolve's zero-units guard returned None → weight fell through to price/units (0 in EUR or X). New code: weight is Amount(0, USD) — same magnitude, different currency bucket for a zero. This matches the residual validator's bucketing (and removes a case where the old currency_accounts path could inject a neutralizing posting the Late balance check then rejected on a valid ledger).
• Correction: the "currency-less specs fall through" bullet describes behavior that was ALREADY true on main (resolve starts with self.currency.clone()?) — it's a preserved invariant, not a change.

New deps: rustledger-query and rustledger-plugin now depend on rustledger-booking (both already sit above it in the build graph via the loader; no cycles).

Testing

• New: test_postings_table_weight_uses_preserved_total pins SELECT weight = exactly 100.00 USD for the PerUnitFromTotal shape (pre-fix: 99.99…); booking unit tests cover all four CostNumber variants and the @/@@ sign rules including zero units.
• Existing booking (110), query (401 integration + parity), and plugin (131) suites all green — the residual arithmetic is byte-identical (same generic functions), and the BQL compat: weight() sign flip on total_annotation.beancount (1 mismatch) #1052 sign tests still pass through the delegation.
• clippy --all-features --all-targets -D warnings, fmt, and cargo doc --all-features -D warnings clean.

🤖 Generated with Claude Code