Silly loaded question here. The bash completions have been sitting on my "need to release" pile for a while now, because there are a bunch of discussions around how/whether to export, how to handle each repo separately, etc etc. I think there is an adequate use case to experiment with dynamically creating completions on tab (which I think Jacob suggested months ago). Solves all kinds of problems with per-repo differences, detecting updates as they're made, etc. However, on my machine that might mean 1-2 seconds lag after tab. Maybe less on M1s (and my soon-to-be, M2).

The scie-pants binary saves you ~33% startup overhead - ~200ms for me. You still have ~500ms overhead in Pants just to noop.

You'd want to generate a cache for help-all that only gets re-genned if plugins added, etc.

Yeah, essentially - basically changes on

pants.toml

touch

Yeah.

I just checked, approx 250ms for ./pants to pants for me - lines up with my tests from the past

The big problem is that gathering options has to import the world and importing is slow.

Ah, okay - that's good to know, also makes more sense