APP_DEBUG=0 in .env means doctrine.dbal.profiling, which defaults to
%kernel.debug%, was off in tests too, so doctrine.debug_data_holder was never
registered. Every test calling countQueries() errored out — all four N+1
regression tests had been dead for as long as they have existed. Turning
profiling on for when@test brings the harness back.
Three of the four passed immediately. The fourth was a real N+1: the
service-coverage endpoint batch-fetched its ServiceItem entities to avoid one
find() per row, but ServiceItem maps staffMembers as fetch: EAGER, so hydrating
N items fired N extra collection loads and the batch bought nothing. Six
coverage rows cost 11 queries where one row cost 6.
ServiceItemRepository::findUuidsByIds() returns the id => uuid map as a scalar
query, so no entity is hydrated and no eager collection is touched.
Also adds the query-count assertion for next_available_at that could not be
written while the harness was broken. Confirmed it fails against the previous
per-day implementation (40 queries for 2 locations, 113 for 6) and passes now.
Suite: 411 tests, 2 failures — both pre-existing and unrelated
(LowTierFixesTest, PatientWalletSessionSettleTest).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
next_available_at called getAvailableSlots() once per day for up to 30 days, and
that helper re-read the schedule, holidays and overrides on every call and then
issued an isSlotTaken() query per candidate slot. Cost grew with both the days
scanned and the slots per day, multiplied by the number of locations.
findNextAvailableStart() fetches the schedule, holidays, overrides and blocking
intervals once for the whole window and walks the days in memory.
Measured on the dev data (a doctor with two locations, first opening several
days out): 73 -> 20 queries for one request. The gap widens as locations or the
distance to the first opening grow.
Reserve appointments must keep blocking here: findBusyIntervals() filters
isReserve = false, so reusing it would have reported a reserved slot as free.
Added findBlockingIntervals(), which mirrors isSlotTaken()'s predicate, and
factored both onto a shared builder.
Verified the endpoint returns identical timestamps before and after.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
