Count installed Scoop packages if available

[FantasyTeddy]

This change adds support for the Scoop installer on Windows.

It works by querying the scoop export command that lists all the installed packages and we can count the number of lines.

[grtcdr]

Lovely stuff... but I'm sure there's a directory whose entries we can count, right?

[FantasyTeddy]

Well... yes I think we could, but I don't know how to determine the correct path to Scoop itself. The default path is under C:\Users\<user>\scoop but it is possible to change this path during installation and if this is the case I'm not sure how to find the correct path...

[grtcdr]

Okay, how are the benchmarks looking?

[FantasyTeddy]

I tried to run the following benchmarks on my machine:	Note	Average	Min	Max
Before this PR	61.8 ms	56.8 ms	114.6 ms
PR without Scoop installed	69.3 ms	57.8 ms	166.8 ms
PR with Scoop installed	65.5 ms	57.4 ms	133.9 ms

[grtcdr]

Thanks for the detail stats.

I can't help but notice that something is very off with these numbers, last time (a little too long ago) we benchmarked macchina, it didn't even exceed 10ms on Windows, but now it's hitting 60ms?...

Are you benchmarking cargo run or plain macchina (the binary)? Because, the former is VERY slow.

[FantasyTeddy]

Oh no, I just realized a mistake I made: even though I remembered to build macchina in release mode, I ran the following command to time it:

Measure-Command { .\target\debug\macchina.exe }

Can you spot my mistake?

---

I ran the tests again with the correct binary and now the results look very bad for my PR:	Note	Average	Min	Max
Before this PR	45.5 ms	40.9 ms	57.9 ms
PR without Scoop installed	601.6 ms	525.4 ms	747.0 ms
PR with Scoop installed	920.1 ms	744.5 ms	1178.7 ms

I totally agree that this is not acceptable and I will have to find a way to determine the installation path of Scoop and do the directory enumeration.

[FantasyTeddy]

I tried to do it that way now: I'm looking for scoop with the which crate and find the correct path to the installed apps relative to this location.

The numbers look much better now, however there is still some impact (especially if the command for scoop cannot be found).	Note	Average	Min	Max
Before this PR	45.5 ms	40.9 ms	57.9 ms
PR without Scoop installed	68.4 ms	51.6 ms	174.4 ms
PR with Scoop installed	53.4 ms	49.5 ms	82.9 ms

[grtcdr]

Can you spot my mistake?

Measure-Command { .\target**debug**\macchina.exe }

Yep.

The tests still show higher numbers than what is expected, can you please try benchmarking but with hyperfine instead? That's what we use for all our benchmarks.

If the scoop command can't be found, then it's safe to assume that no such tool exists, as anyone using scoop will most definitely add it to their PATH.

[FantasyTeddy]

I measured it again with hyperfine and got these results:

Note	Mean [ms]	Min [ms]	Max [ms]	Relative
Latest release (6.0.1)	47.4 ± 3.1	42.7	62.6	1.00
Latest main	43.3 ± 1.1	41.0	46.4	1.00
PR without Scoop installed	54.0 ± 0.8	52.6	55.7	1.00
PR with Scoop installed	52.3 ± 1.7	49.5	60.2	1.00

I'm running an Intel® Core™ i7-4770 CPU @ 3.40GHz so it's not the fastest machine.

[rashil2000]

Sorry for bumping a merged thread, but a few days ago I wanted to add Scoop readout to macchina but couldn't get to it. Glad to see it's landed :)

I'm looking for scoop with the which crate and find the correct path to the installed apps relative to this location.

How much cost does this dependency incur? I think we don't need it.

Well... yes I think we could, but I don't know how to determine the correct path to Scoop itself. The default path is under C:\Users\<user>\scoop but it is possible to change this path during installation and if this is the case I'm not sure how to find the correct path...

The environment variable SCOOP is what you're looking for. See https://github.com/ScoopInstaller/Scoop#install-scoop-to-a-custom-directory-by-changing-scoop

If this variable is unset, Scoop defaults to C:\Users\<user>\scoop.

So if we can check this variable, we can slightly speed things up (I'm assuming an env var lookup is cheaper than searching for an executable in the PATH using a which function).

[grtcdr]

Thank you so much @rashil2000 for this valuable info :heart:

I'm assuming an env var lookup is cheaper than searching for an executable in the PATH using a which function

Absolutely, which (no pun intended) is why LinuxPackageReadout no longer makes more calls than I can count to our own extra::which. I benchmarked the extra::which method and realized it cost around 4300ns for every call. A call to the which::which I would assume is just as slow. So, if we can avoid it, we really should.

@FantasyTeddy care to rewrite it?

[FantasyTeddy]

The environment variable SCOOP is what you're looking for. See https://github.com/ScoopInstaller/Scoop#install-scoop-to-a-custom-directory-by-changing-scoop

If this variable is unset, Scoop defaults to C:\Users\<user>\scoop.

I was looking at this too, but as far as I understand (and I could be totally wrong here) the SCOOP variable is only set temporarily to configure the installation, but it does not persist. However I honestly did not try it and maybe it would work that way and then I am sure this would be a much faster alternative. If I find some time I will happily try this and if it works I will rewrite it.

[rashil2000]

I was looking at this too, but as far as I understand (and I could be totally wrong here) the SCOOP variable is only set temporarily to configure the installation, but it does not persist.

If you see the PowerShell command there:

[Environment]::SetEnvironmentVariable('SCOOP', $env:SCOOP, 'User')

this commands sets environment variables permanently (as opposed to just the current shell session, like on unix).

[FantasyTeddy]

Ah that's the piece of information that was missing for me. Thank you very much for enlightening me :) I will change the implementation accordingly and open another PR.