Grouped content-fingerprint SQL

One pass, no sort and no spill — the whole point, since these tables run to 1.2 billion rows (measured: 325M rows in 7.8 s, so a full release hashes in under a minute).

Usage

content_hash_sql(from, by, cols = c("cell_id", "val"))

Arguments

from

table name or read_parquet(...) expression

by

grouping column — the per-release model id (mdl_seq or mdl_id)

cols

payload columns to hash, in a FIXED order

Value

a SQL string yielding by, n, x, s