indicatif's defaults bottleneck the application on terminal output

[untitaker]

default usage of indicatif currently has horrible performance when many items are processed in short time. there's issues like #170 which say this is due to global locks, but I found that even set_draw_rate(1000) can improve performance by an order of magnitude for certain applications.

I propose that indicatif picks better defaults for draw rate even if it's still printing to the display every millisecond, because the current default experience is broken IMO. Even if it goes as far as enabling steady tick by default to compensate for potential side effects of draw rate, I think spawning an extra thread by default would still be a preferrable default behavior over bottlenecking the entire application on terminal output.

[djc]

Have you tried this with main? There's a bunch of performance improvements.

[djc]

(I do think it might make sense to default to a draw rate of, say, 100. Want to submit a PR?)

[untitaker]

I can will try main branch but I think unless the draw rate default changed or indicatif got significantly faster in tick, there's no way it would change anything

[untitaker]

Very interesting, it seems that indicatif mostly removed whatever bottleneck drawing on the terminal had on master. I can still see a slight improvement with set_draw_rate instead of drawing on every update, but it's more like 10% overhead instead of 2000% overhead

[djc]

It's probably #319, or maybe #361.

[untitaker]

(I do think it might make sense to default to a draw rate of, say, 100. Want to submit a PR?)

there's unresolved questions because "draw rate" currently takes precedence over "draw delta". With a default for "draw rate", draw delta... cannot be set anymore? Unsure.

[djc]

How about you make draw_rate an Option<u64> that defaults to Some(100) and is overwritten with None if the draw_delta is set? (And change the documentation accordingly.)

[djc]

I've implemented a default draw rate in 9df7ab3.

[untitaker]

Thanks! Sorry, didn't get around to contributing this

[djc]

It's fine. My initial fix seems to have caused some regressions, so still investigating some follow up.