Calling process: Query more parents, disable full scan

[th1000s]

Also look at grandparent process when determining the caller

Disable full process scans: The env var DELTA_CALLING_PROCESS_QUERY_ALL re-enables this
last resort method. However usually this is just an expensive scan which doesn't find the caller anyhow.

[th1000s]

The full process scan could still be done on macOS and Windows, I don't think listing all processes requires thousands of syscalls there.

[dandavison]

I've updated @infokiller's docker script to benchmark this branch against 0.11.3 and 0.12.0 (see #977) and the results look good for this branch:

Benchmarking version: 0.11.3
delta 0.11.3
Running command "git diff --no-index a b | delta-0.11.3 >| delta-0.11.3-result" 10 times
61.40 ms per run (614 ms total)
Benchmarking version: 0.12.0
delta 0.12.0
Running command "git diff --no-index a b | delta-0.12.0 >| delta-0.12.0-result" 10 times
59.70 ms per run (597 ms total)
Benchmarking version: dev
delta 0.12.0
Running command "git diff --no-index a b | delta-dev >| delta-dev-result" 10 times
58.30 ms per run (583 ms total)
Spinning up 5000 processes
Process counts:
      1 bash
      1 ps
   5000 sleep
      1 sort
      1 uniq
Benchmarking version: 0.11.3
delta 0.11.3
Running command "git diff --no-index a b | delta-0.11.3 >| delta-0.11.3-result" 10 times
155.20 ms per run (1552 ms total)
Benchmarking version: 0.12.0
delta 0.12.0
Running command "git diff --no-index a b | delta-0.12.0 >| delta-0.12.0-result" 10 times
345.90 ms per run (3459 ms total)
Benchmarking version: dev
delta 0.12.0
Running command "git diff --no-index a b | delta-dev >| delta-dev-result" 10 times
68.80 ms per run (688 ms total)

[dandavison]

Better to use std::env::var here?

[th1000s]

I just never bothered to look up the equivalent for the var() call, which is .map_or(false, ...), until now :)

And I think the underlying getenv() has to walk the environ list just the same.

[dandavison]

Looks great @th1000s! One nitpick (and one clippy complaint).

[dandavison]

Sorry the clippy error was nothing to do with this PR; I've rebased on master to get rid of it.

I am curious whether you prefer std::env::vars over the "get" API std::env::var(name) (two places in process.rs)

[dandavison]

Also, this PR is in Draft status: could you let me know whether it's ready to go out?

(I'm quite keen to make a release soon with the new bat assets, and would like to include this)

[th1000s]

Because I can't test the behavior elsewhere I have only disabled the full scan on Linux now. Other OSes probably have faster syscalls for this query anyhow.

clippy

The 1.58 clippy wants to undo what the (I presume) 1.59 one suggested!

[dandavison]

Thanks very much @th1000s!