fix(gatsby): Use xstate predictableActionArguments & update to 4.34

[LekoArts]

Description

Since statelyai/xstate#3289 there's a warning printed to the console now to use this option. Trying this out if tests break 🤷

Also works around removal of meta.state property mentioned in:

Please be aware that you might not able to use state from the meta argument when using this flag.

Related Issues

[ch54148]

[pieh]

@Andarist maybe you can help us out on another one :)

gatsby/packages/gatsby/src/state-machines/query-running/index.ts

Lines 63 to 81 in 28cfade

	calculatingDirtyQueries: {
	entry: assign<IQueryRunningContext>(({ pendingQueryRuns }) => {
	return {
	pendingQueryRuns: new Set(),
	currentlyHandledPendingQueryRuns: pendingQueryRuns,
	}
	}),
	invoke: {
	id: `calculating-dirty-queries`,
	src: `calculateDirtyQueries`,
	onDone: {
	target: `runningStaticQueries`,
	actions: [
	`assignDirtyQueries`,
	`clearCurrentlyHandledPendingQueryRuns`,
	],
	},
	},
	},

When predictableActionArguments: true is used the order of things changes on this state.

Before it would first do entry / assign and then invoke calculateDirtyQueries service, but with changes in this PR, it first invoke calculateDirtyQueries service and then does entry / assign - is this expected?

The setup for this is done to grab current batch list (pendingQueryRuns) and pass it to service, while reseting it globally so if new entry would show up while we are in service it wouldn't get lost (just would need to wait for another for its turn / next batch execution) - is there better setup to achieve something like that?

[Andarist]

This is essentially a byproduct of how those are concatenated here. I could reverse this but then it wouldn't be possible to do this:

entry: send({ type: 'PING' }, { to: 'child' }),
invoke: {
  id: 'child',
  src: childMachine
}

We could introduce more custom logic for this case to fix this though as I think it makes sense in this case for invoke to have access to the updated context.

I need to think about the potential implementation in v4 as this is a little bit tricky - it doesn't help that the upcoming v5 has much better architecture internally, with which this wouldn't be a problem at all 😬

[pieh]

We could introduce more custom logic for this case to fix this though as I think it makes sense in this case for invoke to have access to the updated context.

I need to think about the potential implementation in v4 as this is a little bit tricky - it doesn't help that the upcoming v5 has much better architecture internally, with which this wouldn't be a problem at all 😬

Please, do let us know wether we should expect changes in this behavior. Maybe consider temporarily disabling the warning that is printed to console about predectableActionArguments until there is clarity on what the behavior of that mode should be? I do think in meantime we will pin xstate to previous minor, as our users are seeing those warnings which we are trying to "silence" with this PR

[Andarist]

Please, do let us know wether we should expect changes in this behavior. Maybe consider temporarily disabling the warning that is printed to console about predectableActionArguments until there is clarity on what the behavior of that mode should be?

I think that I've figured out a potential way forward yesterday. I need to give a try implementing it but I'm hopeful - what you have discovered here was not an intentional change, so I very much would like to fix this for you and other users.

I do think in meantime we will pin xstate to previous minor, as our users are seeing those warnings which we are trying to "silence" with this PR

That's a good move for you right now - I won't be able to push a fix today and I don't want to turn off the warning just to enable it again in a couple of days.

[LekoArts]

@Andarist heh, it's been a while... I just merged your PRs and update to latest xstate, let' see if tests are happy

[Andarist]

Let me know if you'd like any further assistance with this.

[LekoArts]

I initially merged #36393 but that failed some tests and for now I'm fine with it being not optimal but working :D

Appreciate your input and help!