Fix device selection

[tokou]

Proposed changes

• The issue seems to be in the logic related to selecting the devices depending on the shards
• I tried cleaning it up as best as I could but that part of the code is kinda all over the place
• I started working on launching existing devices in addition to creating new ones for shards but I wonder what's the value from such a feature

Testing

• Launch 2 devices
• Run a folder with multiple flows without specifying sharding
• Run a folder with multiple flows with shards = 2
• Run a folder with multiple flows with shards = 3

Issues fixed

Fixes #1849

[bartekpacia]

Another false positive of #2005...

[bartekpacia]

I tested this and it seems the issue is not fixed, i.e. Maestro still doesn't ask what device to run on when more than 1 device is connected.

To reproduce, start 2 android emulators, and then:

./maestro-cli/build/install/maestro/bin/maestro test --include-tags passing,failing --exclude-tags ios e2e/workspaces/demo_app

[bartekpacia]

Aside from that it looks good. Also thanks a ton for resolving all merge conflicts!

[bartekpacia]

Ugh, more conflicts after I merged #1995

[bartekpacia]

I started working on launching existing devices in addition to creating new ones for shards but I wonder what's the value from such a feature

I think it's OK to not do this.

Actually, I don't like that Maestro CLI can create new devices as part of maestro test. In my opinion, to keep code cleaner and responsibilities separate, maestro test should merely print "Not enough devices. Start some or create with maestro start-device" instead of offering the user to go through the device creation flow.

[tokou]

Yeah I fully agree
But the feature was present that's why I kept it. I'll gladly get rid of that part of the code

[tokou]

@bartekpacia I got rid of device creation. I also switched --shard-all to a boolean as it makes more sense to me like that. Let me know what you think.

[bartekpacia]

This is awesome. Thank you.

Re: --shard-all – I prefer to keep it as an int arg, for (1) consistency with --shard-split, and also I think it's just useful to expose it to users and doesn't cost us much.

[DavidREntwistle]

This fix doesn't look to have resolved issue in CLI v. 1.39.0.
I've re-raise this issue now here: #2096