[WC-3533]: Fix/combobox on change

[r0b1n]

Pull request type

Refactoring (e.g. file rename, variable rename, etc.)

---

Description

Previously we were calling the event defined as OnChange seprately from the attributes and associations. In the code we were calling it manually to emulate the framework behaviour because of some istorical reasons. At this moment analysis says that we can actually attach the onChange to attributes and association and make the framework handle this action automatically.

This is needed to mitigate the issue with a race condition between editing and calling actions. At the moment if association being edited is also part of the onCHange action parameters, the action (nanoflow) gets stale parameters, so when changing from Options 1 to Option 2 and then immediately calling manual onChange, the nanoflow gets Option 1, while user just changed Option 1 to Option 2.

This refactoring passes the responsibility of handing the race condition to the framework's built in functionality.

What is still broken

As I implemented this change, this revealed a possible bug in the framework where instead of stale parameter, the prameter is not passed at all. 😅 Waiting for the team to confirm if it's a bug on their side. For now one e2e test is expected to be broken.

What should be covered while testing?

Check that all existing OnChange actions are still working as expected for different types of attributes and association. This PR introduces e2e tests that cover majority of scenarios, there might be some I missed. Test pages are committed in the test project and stil lcompatible with 10.22, so you can check with old version of the widget.

[github-actions]

AI Code Review

⚠️ Approved with suggestions — low-severity items only, safe to merge

---

What was reviewed

File	Change
src/Combobox.xml	Adds onChange="onChangeEvent" to 5 attribute/association properties
typings/ComboboxProps.d.ts	Removes onChangeEvent?: ActionValue from ComboboxContainerProps (now platform-managed)
src/helpers/Association/BaseAssociationSelector.ts	Removes onChangeEvent field and executeAction call from base setValue
src/helpers/Association/AssociationMultiSelector.ts	Removes super.setValue(value) call (was the only caller of executeAction)
src/helpers/Association/AssociationSingleSelector.ts	Removes super.setValue(value)
src/helpers/Association/utils.ts	Removes onChangeEvent from extraction tuple
src/helpers/Database/DatabaseSingleSelectionSelector.ts	Removes onChangeEvent, _valuesIsEqual guard, and executeAction call
src/helpers/Database/utils.ts	Removes onChangeEvent from ExtractionReturnValue
src/helpers/EnumBool/EnumBoolSingleSelector.tsx	Removes onChangeEvent field and executeAction call
src/helpers/Static/StaticSingleSelector.ts	Removes onChangeEvent field and executeAction call
src/helpers/Static/utils.ts	Removes onChangeEvent from extraction tuple
e2e/OnChange.spec.js	New E2E spec covering all 6 source types + 1 fixme for known framework bug
e2e/utils/Combobox.pageObject.js	New page object for combobox interaction
e2e/utils/logEntryParser.js	New parser for change-tracking microflow log output
CHANGELOG.md	Adds "improved reliability of On change action" under Unreleased / Fixed
docs/requirements/backend-structure.md	Documents platform behaviour for onChange-bound actions
openspec/changes/archive/…	Archived change artifacts (design, proposal, spec, tasks)

Skipped (out of scope): openspec/ spec files, pnpm-lock.yaml, dist/

---

Findings

⚠️ Low — waitFrames(page, 10) is a frame-count timing wait

File: e2e/OnChange.spec.js line 127
Note: waitFrames(page, 10) pauses for 10 render frames before reading the log text. It works in practice but is fragile under load — if the microflow takes longer than expected the assertion can see stale output. A web-first assertion on the text widget first would make this resilient:

async function getLogs(page) {
    // wait for at least one log entry to appear before reading
    await expect(page.locator(".mx-name-text2")).not.toBeEmpty();
    const text = await page.locator(".mx-name-text2").innerText();
    return parseLogEntries(text);
}

Not blocking — waitFrames is used elsewhere in the repo and works — but worth improving before adding more tests that rely on this helper.

---

⚠️ Low — logEntryParser.js comment says "comma-separated" but the separator is !

File: e2e/utils/logEntryParser.js line 279
Note: The JSDoc above LIST_FIELDS reads "Field names whose raw value is a comma-separated list" but coerceListValue splits on "!". The block comment earlier in the file correctly describes the !-prefix format. A one-word fix prevents confusion for anyone reading just the inline JSDoc.

/** Field names whose raw value is a "!"-separated list (multi-relation). */
const LIST_FIELDS = new Set(["multiAssocTitles"]);

---

Positives

• test.describe.fixme used correctly to quarantine the known-broken "pass association to onChange" test — the bug is in the framework, not the widget, and the comment explains why.
• docs/requirements/backend-structure.md updated in the same PR to document the platform behaviour, so future reviewers won't mistake the absent onChangeEvent prop for a typings sync issue.
• The _valuesIsEqual guard removal is explicitly justified in the design doc — the platform already deduplicates by value, so the guard was dead code.
• Boilerplate executeAction calls cleanly removed from every selector in one shot; the openspec change documents each touched file, making the scope easy to audit.
• CHANGELOG entry is user-facing and accurate ("reliability of On change action") without exposing internal refactoring details.