Sqlx: 0.4 performance decrease

[sanpii]

Hello,

I wrote a benchmark to help me in the development of my postgresql crate, but the result of the update to 0.4 may interess you: https://github.com/elephantry/benchmark/commit/1355a95a5770118f6c5764da8e087020287f7445?short_path=3b8c4a9#diff-3b8c4a9f770124569b674a59525322ea77fb31bd9ef1e6a8edb68639e9ecec5b

The corresponding code:

sqlx::query_as::<_, User>("SELECT id, name, hair_color, created_at, null as posts FROM users").fetch_all(self)

[OskarPersson]

I've seen the same kind of decrease in my project since upgrading from 0.3.5 to 0.4.1. I will try to create a small reproducible version of it later to showcase it

[OskarPersson]

I tried to do some git bisecting using the benchmark tool that @sanpii referred to. I had to skip some commits because of compilation error and I found that https://github.com/launchbadge/sqlx/commit/9b68eb19 is the first bad commit that I could run and https://github.com/launchbadge/sqlx/commit/a233fbfd is the last good commit that I could run:

https://github.com/launchbadge/sqlx/commit/a233fbfd:

test sqlx::query_all       ... bench:   9,357,750 ns/iter (+/- 1,016,965)

https://github.com/launchbadge/sqlx/commit/9b68eb19:

test sqlx::query_all       ... bench:  24,725,440 ns/iter (+/- 19,397,179)

[mehcode]

I really appreciate all of this. Performance is going to become a priority for 0.5 and beyond. Currently we've paid very little attention to how fast the driver is.

Self-contained benchmarks that we can run on 0.3, 0.4, and soon 0.5 will help. I'm less interested at this time for how it runs compared to another postgres crate.