[enghitalo]

Gains around 20% for now in vlib/v/tests/bench/bench_json_vs_json2.v;

benchmark to verify impact when using push_many

import time
import benchmark

const max_iterations = 100_000

fn main() {
	value := 'abcdefghijklmnopqrstuvdadkaodkpaiosdjaodijaodiajodia'

	mut buffer := []u8{}

	mut cap_inited_buf := []u8{cap: value.len * max_iterations}

	mut buffer2 := []u8{}

	mut cap_inited_buf2 := []u8{cap: value.len * max_iterations}

	mut b := benchmark.start()

	for i := 0; i < max_iterations; i++ {
		unsafe {buffer.push_many(value.str, value.len)}
	}

	b.measure("unsafe {buffer.push_many(value.str, value.len)}")

 	for i := 0; i < max_iterations; i++ {
		for letter in value {
			cap_inited_buf  << letter
		}
	}

	b.measure("cap_inited_buf  << letter")

	for i := 0; i < max_iterations; i++ {
		for letter in value {
			buffer2  << letter
		}
	}

	b.measure("buffer2  << letter")

	for i := 0; i < max_iterations; i++ {
		unsafe {cap_inited_buf2.push_many(value.str, value.len)}
	}

	b.measure("unsafe {cap_inited_buf2.push_many(value.str, value.len)}")
}

• Aims: Use recursion, at compile time, without using workarounds.

• option
• sumtype
• alias
• pointer
• enum
• array

TODO

• open a issue about chars_in_struct that should be working (This is breaking the CI)

[spytheman]

The new Count struct and its related API, does not need to be public.

The tests for it are very repetitive. The test file can also be an internal one (starting with module json2), thus avoiding exposing the helper Count struct.

[enghitalo]

The new Count struct and its related API, does not need to be public.

The tests for it are very repetitive. The test file can also be an internal one (starting with module json2), thus avoiding exposing the helper Count struct.

I thought this might be useful in cases like

mut count := json.Count{0}

count.count_chars(val)

mut request_buffer := 'POST / HTTP/1.1\r\nContent-Length: ${count.get_total()}\r\n\r\n'.bytes()

request_buffer.ensure_cap( request_buffer.len + count.get_total())

encoder.encode_value(val, mut request_buffer)!

What do you think?

[spytheman]

I think, that we already have io reader/writer interfaces.

I also think that making something public, if there is a need for it in the future, justified by actual code and actual measurements in concrete situations, is easy.

In contrast, turning a public API to a private one, is much harder, costing time for deprecation, to minimize the breaking change. I also think, that the cost for both maintainers and users, of a non essential, inconvenient API is high.

[enghitalo]

You rock! Thanks for the review.