stephenafamo / scan / 28873360656

90%

main: 90%

Pull #9

github

perf: reuse per-row scan buffers in Row

The per-row scan path (scanCurrentRow, run once per row for every mapper
type and every query) allocated two slices on each row:

- scanDestinations was reallocated with make() purely to reset state,
  even though the before hook re-populates the same indices
  deterministically each row.
- createTargets allocated a fresh []any for rows.Scan on every row,
  even though drivers don't retain it past the Scan call.

Keep both slices on Row, allocated once in wrapRows, and reuse them each
row: zero scanDestinations in place (go.mod is go 1.18, so no clear()
builtin) and refill scanTargets — every index is overwritten each row,
so no stale values linger.

Safe because before(v) -> scanCurrentRow -> after(val) is synchronous
per row and columns are constant per query.

Wide-struct bench (11 cols, 100 rows):
  ScanAll: 520 -> 321 allocs/op (-38%), 134704 -> 88512 B/op (-34%)
  ScanOne:  16 -> 15 allocs/op,          2808 ->  2552 B/op

5 of 5 new or added lines in 1 file covered. (100.0%)

577 of 641 relevant lines covered (90.02%)

42.4 hits per line