Builds the per-unit batching clients of one transport, and the cache they
share.
Declaring a client and reaching for one are separate operations, because every
option a client takes is a fact about the unit rather than about a caller: two
batching clients on one unit hold two batches and coalesce neither, two caches
on one unit hold two beliefs about one device, and one batch has one window. A
per-caller configuration could only ever be correct when every caller passed
the same value, so it is declared once instead of compared at every call.
withBatchingClient declares. batchingClient looks up, and waits for a
declaration already under way rather than depending on which fiber ran first.
Builds the per-unit batching clients of one transport, and the cache they share.
Declaring a client and reaching for one are separate operations, because every option a client takes is a fact about the unit rather than about a caller: two batching clients on one unit hold two batches and coalesce neither, two caches on one unit hold two beliefs about one device, and one batch has one window. A per-caller configuration could only ever be correct when every caller passed the same value, so it is declared once instead of compared at every call.
withBatchingClientdeclares.batchingClientlooks up, and waits for a declaration already under way rather than depending on which fiber ran first.