Alternative approach:
synchronous hostUpdated() instead of async fallback
The current fix adds a fallback in getSlotted() to query slot elements directly when #slotRecords hasn't been populated yet. This addresses the symptom but leaves the root cause: hostConnected and #initSlotMap are async, creating a timing gap where no data source exists.
The gap also affects hasSlotted() and isEmpty(), which aren't addressed here.
Why hostConnected is async
Both hostConnected and #initSlotMap await host.updateComplete just to ensure the shadow DOM exists so they can query for elements. But Lit already provides a synchronous hook that guarantees this: hostUpdated(), which runs after every render when the DOM has been committed.
Suggested change
Make hostConnected synchronous (just sets up the MutationObserver), move record initialization to hostUpdated(), and have the MutationObserver trigger a re-render instead of calling #initSlotMap directly:
hostConnected(): void {
this.#mo.observe(this.host, { childList: true });
this.#slotRecords.clear();
}
hostUpdated(): void {
if (this.#slotRecords.size === 0) {
for (let slotName of this.#slotNames.concat(Object.values(this.#deprecations))) {
slotName ||= SlotController.default;
const slot = this.#getSlotElement(slotName);
if (slot) {
this.#slotRecords.set(slotName, new SlotRecord(slot, slotName, this.host));
}
}
if (this.#slotRecords.size > 0) {
this.host.requestUpdate();
}
}
}
The MutationObserver callback becomes:
#mo = new MutationObserver(() => this.host.requestUpdate());
And #initSlotMap can be removed entirely.
This also requires updating the SlotControllerPublicAPI declaration to match:
Why this works
- First render: Records empty,
isEmpty() returns true (matches SSR server behavior). hostUpdated() populates records and calls requestUpdate().
- Second render: Records exist,
isEmpty()/getSlotted() read live DOM via SlotRecord getters, correct results. hostUpdated() sees records exist, no requestUpdate(), no infinite loop.
- MutationObserver: Child changes trigger
requestUpdate() → re-render → SlotRecord getters reflect new DOM state. No record rebuild needed since they query live DOM.
- Reconnection:
hostConnected clears records → next render → hostUpdated() repopulates.
What this fixes that the current PR doesn't
hasSlotted() and isEmpty() have the same timing gap, they also read from #slotRecords and return incorrect results before async init completes. The hostUpdated() approach fixes all three methods.
- Eliminates the
async architecture that caused the issue rather than working around it.
Trade-off
The two-render pattern on initial connection is inherent <slot> elements are created by render(), so the first render can't know about them. This was already the case with the async approach.
We can spin this off to another issue if we feel the need
Originally posted by @zeroedin in #3093 (comment)
Alternative approach:
synchronous
hostUpdated()instead of async fallbackThe current fix adds a fallback in
getSlotted()to query slot elements directly when#slotRecordshasn't been populated yet. This addresses the symptom but leaves the root cause:hostConnectedand#initSlotMapareasync, creating a timing gap where no data source exists.The gap also affects
hasSlotted()andisEmpty(), which aren't addressed here.Why
hostConnectedisasyncBoth
hostConnectedand#initSlotMapawait host.updateCompletejust to ensure the shadow DOM exists so they can query for elements. But Lit already provides a synchronous hook that guarantees this:hostUpdated(), which runs after every render when the DOM has been committed.Suggested change
Make
hostConnectedsynchronous (just sets up the MutationObserver), move record initialization tohostUpdated(), and have the MutationObserver trigger a re-render instead of calling#initSlotMapdirectly:The MutationObserver callback becomes:
And
#initSlotMapcan be removed entirely.This also requires updating the SlotControllerPublicAPI declaration to match:
Why this works
isEmpty()returns true (matches SSR server behavior).hostUpdated()populates records and calls requestUpdate().isEmpty()/getSlotted()read live DOM via SlotRecord getters, correct results.hostUpdated()sees records exist, norequestUpdate(), no infinite loop.requestUpdate()→ re-render → SlotRecord getters reflect new DOM state. No record rebuild needed since they query live DOM.hostConnectedclears records → next render →hostUpdated()repopulates.What this fixes that the current PR doesn't
hasSlotted()andisEmpty()have the same timing gap, they also read from #slotRecords and return incorrect results before async init completes. ThehostUpdated()approach fixes all three methods.asyncarchitecture that caused the issue rather than working around it.Trade-off
The two-render pattern on initial connection is inherent
<slot>elements are created byrender(), so the first render can't know about them. This was already the case with theasyncapproach.We can spin this off to another issue if we feel the need
Originally posted by @zeroedin in #3093 (comment)