Submit a pull request

Linear: FLU-

Github Issue: #

CLA

• I have signed the Stream CLA (required).
• The code changes follow best practices
• Code changes are tested (add some information if not applicable)

Description of the pull request

Bug

_MessageCardState._updateWidthLimit is scheduled via WidgetsBinding.instance.addPostFrameCallback from didChangeDependencies, and reads renderBox.size.width when it fires:

void _updateWidthLimit() {
  final attachmentContext = attachmentsKey.currentContext;
  final renderBox = attachmentContext?.findRenderObject() as RenderBox?;
  final attachmentsWidth = renderBox?.size.width;          // <-- throws
  ...
}

If the attachments subtree has been detached from the tree between scheduling and firing (e.g. the parent rebuilt the list, the chat was disposed, the attachment was deleted, the user navigated away), then per Flutter's contract RenderBox.size throws StateError: RenderBox was not laid out:

A RenderBox object must be laid out before any method that reads its size or position is called.

The throw is caught by FlutterError.onError, so the user sees a dropped frame rather than a crash. But it surfaces as a fatal error in crash reporters like Sentry. In a production Flutter app pinned to stream_chat_flutter: ^9.24.0, the two Sentry issues corresponding to this code path have generated over 1500 events across 1100+ unique users in a single 2-week release — the loudest non-network issue in the audit.

Sample symbolicated stack:

at _MessageCardState.didChangeDependencies.<T> (message_card.dart:155)
at _MessageCardState._updateWidthLimit (message_card.dart:138)
at RenderBox.size (box.dart:2304)
    (throw StateError('RenderBox was not laid out: $runtimeType#${shortHash(this)}'));

Fix

1. Add a hasSize guard before reading renderBox.size — exactly what the framework docs recommend for the "read size after a frame" pattern.
2. Move the existing mounted check from "before setState" to the top of the function. Previously, on an unmounted widget we still touched the (possibly unlaid) render object — only the setState was skipped. The earlier check is symmetrical and slightly cheaper.

Diff:

void _updateWidthLimit() {
  if (!mounted) return;

  final attachmentContext = attachmentsKey.currentContext;
  final renderBox = attachmentContext?.findRenderObject() as RenderBox?;
  // The attachments subtree may have been detached between scheduling this
  // post-frame callback and it firing. Reading [RenderBox.size] without
  // checking [RenderBox.hasSize] throws `RenderBox was not laid out`.
  if (renderBox == null || !renderBox.hasSize) return;
  final attachmentsWidth = renderBox.size.width;

  if (attachmentsWidth == 0) return;

  setState(() => widthLimit = attachmentsWidth);
}

Testing

This is a defensive guard against a framework-level race that fires intermittently on real user devices and is hard to reproduce locally. The guard preserves the existing happy-path behavior (when the render object is laid out, width is read and applied identically to before) and short-circuits cleanly when it's not.

Screenshots / Videos

Not applicable — this fix removes a race-condition error log, no visual change.

Summary by CodeRabbit

• Bug Fixes
  • Fixed a crash when measuring attachment widths if attachments are removed before a post-frame update; width checks now avoid invalid reads and skip updates when not applicable.
• Tests
  • Added widget tests verifying width measurement behavior, unmount safety, zero-width handling, and no-op when no attachments exist.
• Documentation
  • Updated changelog with the fix.