[core] Fix restarted root-level invocations #3073

davidkpiano · 2022-02-22T01:56:27Z

External root-level transitions (which should be invalid – only internal root-level transitions are allowed) will no longer restart root-level invocations. See #3072 for more details.

Fixes #3072

changeset-bot · 2022-02-22T01:56:29Z

🦋 Changeset detected

Latest commit: 9e1d615

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
xstate	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

ghost · 2022-02-22T01:57:39Z

CodeSee Review Map:

Review in an interactive map

View more CodeSee Maps

Legend

codesandbox-ci · 2022-02-22T01:57:40Z

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

Latest deployment of this branch, based on commit 9e1d615:

Sandbox	Source
XState Example Template	Configuration
XState React Template	Configuration
Unable to send event to child	Issue #3072
Unable to send event to child (forked)	Issue #3072

davidkpiano · 2022-03-26T20:01:41Z

packages/core/src/StateNode.ts

+    const reentryNodes: AnyStateNode[] = [];
+
+    if (!isInternal) {
+      allNextStateNodes.forEach((n1) => {


This is the actual change

Andarist · 2022-03-28T08:54:57Z

I've taken a look at this from the SCXML perspective and I think that this is a good change. Let's start first with how this compares to the spec text.

We can read that <scxml> is never an explicit part of the configuration (configuration definition). This could be read as if it is an implicit member of the configuration or just that it can't ever be exited and thus it's not worth keeping it in the configuration.

It's also worth noting that entrySet and exitSet are computed based on "transition domain". Only descendants of the domain can be exited and thus reentered (see the computeExitSet). For external transitions, this is based on the findLCCA procedure that mentions this:

The Least Common Compound Ancestor is the or element s such that s is a proper ancestor of all states on stateList and no descendant of s has this property. Note that there is guaranteed to be such an element since the wrapper element is a common ancestor of all states.

This clearly means that we can't "exit" the root "state". The difference between SCXML and XState though is that in XState we actually extend what the root (<scxml>) can do! In SCXML you can't define <invoke>, <onentry>, nor <onexit> on the <scxml> (those are not valid children for that element).

Ok, so I think we are in the agreement that this is something that we'd like to do. However, I wonder if we really should do this in v4. The current behavior was with us here for quite a long time and it's also not a deal-breaker, it's just something that we'd like to change - but by no means this is a pressing matter.

I know that we've defined the semver policy about breaking changes so we have some freedom to change this now. That doesn't mean though that we should introduce such changes lightly. So my question is - is the RoI of this change sufficiently high that we are OK with this "breaking change" (cause it definitely might be for some state machines out there, it's hardly an edge case)? Isn't this something that we should just change in v5?

Andarist · 2022-03-28T08:56:55Z

packages/core/test/invoke.test.ts

@@ -2905,6 +2905,41 @@ describe('invoke', () => {
      done();
    }, 0);
  });
+
+  // https://github.com/statelyai/xstate/issues/3072
+  it('root invocations should not restart on external transitions', (done) => {


We should also add a separate test that would test, in a more explicit way, that the root state is never being reentered here. That invoke's implementation is piggybacking on entry/exit actions is an implementation detail so I don't think that we should assume that this is also testing that more "basic" scenario.

Andarist · 2022-03-28T09:02:18Z

packages/core/src/StateNode.ts

-      : flatten(allNextStateNodes.map((n) => this.nodesFromChild(n)));
+    const reentryNodes: AnyStateNode[] = [];
+
+    if (!isInternal) {


Note that this algorithm is also broken in other subtle ways. States are not only reentered for external transitions - internal/external only change what states are potentially reentered.

For instance, on this codesandbox this entry action should technically be called twice. Even though the transition is internal, this target state should be reentered.

And I'm not saying that we should fix this right now. I'm just observing the fact. I somewhat don't see how fixing non-critical bugs in v4 gets us closer to anything. As mentioned before - I would rather halt the v4 development (just fix pressing issues) and focus on v5 (or on the "satellite" packages, as that development is mostly orthogonal)

Andarist · 2022-03-28T09:05:52Z

.changeset/four-waves-lick.md

+---
+'xstate': patch
+---
+
+The `AnyStateNode` type, which represents a `StateNode` of any context, event, etc., has been introduced.


This is a somewhat internal change - I'm not sure how relevant this is for our users and I would probably skip "documenting" it:

Suggested change

---

'xstate': patch

---

The `AnyStateNode` type, which represents a `StateNode` of any context, event, etc., has been introduced.

---

---

Andarist · 2022-03-29T12:49:17Z

@davidkpiano I've started merging this into the next branch. I was somewhat surprised that this test has passed there without any changes to the implementation that is on the next branch. So I've decided to recheck if it passes "correctly" or "accidentally".

I've verified that it was passing accidentally and I've recognized 2 issues:

stop+invoke/spawn for the same actor in the same macrostep, wrong actor gets stopped and no actor is present in the state after the transition, this creates a memory leak
getTransitionDomain/findLCCA might return null/undefined and that is wrong - something should always be returned from there.

I'm going to tackle both and then merge this PR to the next branch and fix the linked ticket there.

… part of the #3073

…ansitions (#3180) * Simplify the algorithm for computing reentry nodes during external transitions * Remove the logic preventing root from being reentered that originally part of the #3073

Andarist · 2022-08-01T13:24:17Z

I've created #3487 as a successor PR to this one. The new one is targeting the next branch

davidkpiano added 2 commits February 21, 2022 20:53

Fix restarted root-level invocations

ed9197f

Add changesets

ee4c8ee

davidkpiano requested a review from Andarist February 22, 2022 01:56

davidkpiano added 3 commits February 26, 2022 16:13

Merge branch 'main' into davidkpiano/3072

58a9c7a

Merge branch 'main' into davidkpiano/3072

dcb1e02

Merge branch 'main' into davidkpiano/3072

429eaeb

davidkpiano commented Mar 26, 2022

View reviewed changes

Andarist force-pushed the davidkpiano/3072 branch 2 times, most recently from d52dcb5 to 6a385f4 Compare March 27, 2022 19:11

Andarist reviewed Mar 28, 2022

View reviewed changes

davidkpiano force-pushed the davidkpiano/3072 branch from 6a385f4 to 429eaeb Compare March 28, 2022 15:11

Merge branch 'main' into davidkpiano/3072

9e1d615

Andarist added a commit that referenced this pull request Jun 23, 2022

Remove the logic preventing root from being reentered that originally…

5c0ecae

… part of the #3073

Andarist closed this Aug 1, 2022

Andarist deleted the davidkpiano/3072 branch March 22, 2023 15:05

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[core] Fix restarted root-level invocations #3073

[core] Fix restarted root-level invocations #3073

davidkpiano commented Feb 22, 2022

changeset-bot bot commented Feb 22, 2022 •

edited

Loading

ghost commented Feb 22, 2022 •

edited by ghost

Loading

codesandbox-ci bot commented Feb 22, 2022 •

edited

Loading

davidkpiano Mar 26, 2022

Andarist commented Mar 28, 2022

Andarist Mar 28, 2022

Andarist Mar 28, 2022

Andarist Mar 28, 2022

Andarist commented Mar 29, 2022

Andarist commented Aug 1, 2022

[core] Fix restarted root-level invocations #3073

[core] Fix restarted root-level invocations #3073

Conversation

davidkpiano commented Feb 22, 2022

changeset-bot bot commented Feb 22, 2022 • edited Loading

🦋 Changeset detected

ghost commented Feb 22, 2022 • edited by ghost Loading

CodeSee Review Map:

Legend

codesandbox-ci bot commented Feb 22, 2022 • edited Loading

davidkpiano Mar 26, 2022

Choose a reason for hiding this comment

Andarist commented Mar 28, 2022

Andarist Mar 28, 2022

Choose a reason for hiding this comment

Andarist Mar 28, 2022

Choose a reason for hiding this comment

Andarist Mar 28, 2022

Choose a reason for hiding this comment

Andarist commented Mar 29, 2022

Andarist commented Aug 1, 2022

changeset-bot bot commented Feb 22, 2022 •

edited

Loading

ghost commented Feb 22, 2022 •

edited by ghost

Loading

codesandbox-ci bot commented Feb 22, 2022 •

edited

Loading