Got a benchmark?

I ran a micro-benchmark focused specifically on the result parsing phase, isolating it from network and I/O overhead.

To avoid noise from TCP and libpq I/O, I mocked the pq interface and types.getTypeParser() so the benchmark measures only the cost of:

• building fields
• building rows
• parsing values

The mock generates a result set of 10,000 rows × 10 columns.
Both implementations run against the exact same mocked data and parser logic.

Results (ops/sec):

• PR: ~397 ops/s
• master: ~341 ops/s

That’s roughly a +16% improvement in this isolated scenario.

I’m aware this is not an end-to-end benchmark and does not include network latency or query execution time, but the goal here was to validate that caching parsers removes repeated calls to getTypeParser() and measurably reduces overhead in the hot path of result processing.

see here