another Go postgres driver about pg HOT 7 CLOSED

I have an idea how to improve BenchmarkPgSelectMultipleRowsPrepared and BenchmarkPgSelectMultipleRowsUnprepared (base on fields that bypass a lot reflect work: https://github.com/vmihailenco/msgpack/blob/master/typeinfo.go#L20). BenchmarkRawSelectMultipleRowsPrepared looks like a cheating and TBH I don't understand what it does. The rest of benchmarks looks good :)

from pg.

Well, it's documented that the raw methods are basically cheating in order to determine a lower bound to the attainable results. Pretty clever: https://github.com/jackc/go_db_bench

from pg.

What I found so far is that changing people to:

type People []person

func (people *People) New() interface{} {
    pp := *people
    pp = append(pp, person{})
    return &pp[len(pp)-1]
}

makes some benchmarks almost 30-40% faster. So I guess it is possible to write such benchmarks where go-pg will be much faster...

from pg.

I've actually always wondered why you used slices of pointers instead of just slices. Good to know!

from pg.

Wait, shouldn't it be:

type People []person

func (people *People) New() interface{} {
    p := person{}
    *people = append(*people,p)
    return &p
}

Because otherwise you append to a new slice (which refers to the same memory, which might get reallocated), and the old slice doesn't get updated.

Unless I'm misunderstanding something.

from pg.

Disregard me, that didn't work because it returned a copy of person{}... ugh. This is the corrected version:

*people = append(*people, person{})
dpeople := *people
return &dpeople[len(dpeople)-1]

What do you think?

from pg.

I consider this fixed by jackc/go_db_bench#2. go-pg is still slightly slower, but that is the cost of using reflect.

from pg.