Update: layered Flutter 3.35-3.40 compat patch on top of the merge

Pushed an additional commit (73e427c7) on top of the merge to fix a build break against our pinned Flutter SDK.

What changed

Drop upstream's 3-line @override void updateStyle(TextInputStyle style) from super_editor/lib/src/default_editor/document_ime/ime_decoration.dart. Replace with a documented comment block.

Why this is needed

• Upstream's PR #2950 (commit b422c326) was authored against Flutter master immediately after flutter/flutter#180436 added TextInputConnection.updateStyle(TextInputStyle) and deprecated setStyle(...).
• Per Flutter's breaking-changes page (last updated 2026-03-03), the change is "Landed in version: TBD / In stable release: Not yet."
• Our pinned Flutter is 3.35.7. Its text_input.dart declares only setStyle({...}). class TextInputStyle and TextInputConnection.updateStyle(...) don't exist.
• TextInputConnectionDecorator implements TextInputConnection, so the override fails to compile against any current stable Flutter — the parent doesn't have updateStyle to override, and TextInputStyle is undefined.

Why we can't migrate to updateStyle instead

There's no Dart conditional-imports trick to selectively include an interface override based on the host Flutter version — conditional imports switch entire libraries, not individual class members. The only paths are: (a) drop the override on stable Flutter, (b) wait for the API to ship in stable. Option (a) is what this commit does.

Why this is safe

setStyle(...) immediately above the dropped block is still in the framework and still in the parent interface — it covers the entire surviving public API on stable Flutter. Removing the would-be-override is functionally a no-op (the framework can't call a method that doesn't exist on the parent).

Restore criteria (documented in the code + FORK_CHANGES.md)

1. Flutter stable ships TextInputStyle + TextInputConnection.updateStyle(...).
2. This monorepo's pinned Flutter SDK is upgraded to ≥ that release.
3. Then revert this commit.

Tracked

Added as fork divergence #6 in FORK_CHANGES.md, explicitly marked TEMPORARY. The Pre-Merge Checklist in that file now flags ime_decoration.dart so future syncs catch upstream re-introducing the override.

Diff stat

2 files changed, 50 insertions(+), 3 deletions(-)
  super_editor/lib/src/default_editor/document_ime/ime_decoration.dart
  FORK_CHANGES.md

TL;DR

We want all 40 upstream fixes from the merge + this one tiny compat patch on top to keep the fork buildable on Flutter 3.35.7 until upstream's updateStyle API actually ships in stable.

Fork Divergence Audit + Flutter 3.35-3.40 Compat Investigation

This is a deeper audit done after the merge + compat patch landed. Two parts:

1. Part 1 — Fork Divergence Audit (Desktop-First): every fork PR traced to actual call sites in the desktop client + applets, compared against current upstream behavior at 0.3.0-dev.51, with keep/contribute/replace verdicts.
2. Part 2 — Flutter 3.35-3.40 Compat Investigation: the deeper TextInputStyle issue — fact-checked via Flutter docs + Perplexity, with restoration plan.

Part 1 — Fork Divergence Audit

Headline Verdict

Bottom line: all 5 are still load-bearing on desktop. Nothing to delete. #3 and #4 are upstream-contribution candidates that would shrink the permanent fork delta from 5 → 3.

PR #1 — No IgnorePointer in TextComponent.build()

What it does

Removes upstream's IgnorePointer wrapper around SuperText inside TextComponent.build() so pointer events propagate to ancestor widgets.

Where it's exercised

• frontend/pieces_platform_client_sdk/.../pieces_markdown_reader.dart:681-682 wraps SuperReader in a Listener whose onPointerDown: _handlePointerDown runs:
  • _readerFocusNode.requestFocus() on every pointer down (line 567+)
  • Ctrl/Meta + secondary mouse click → _showContextMenu(...) (lines 567-584) — desktop-specific context menu
• 11 reader embed sites depend on this:
  • frontend/pieces_copilot_module/.../system_message_renderer.dart:319, 494, 584, 1327
  • frontend/pieces_copilot_module/.../copilot_message_summary_content.dart:212
  • frontend/pieces_copilot_module/.../copilot_chat_message_markdown_content.dart:127
  • frontend/pieces_workstream_activity_module/.../workstream_rollup_super_reader_view.dart:49
  • frontend/pieces_workstream_activity_module/.../workstream_editing_copilot_panel.dart:305
  • frontend/pieces_platform_client_sdk/.../copilot_quoted_selection_widget.dart:78
  • frontend/pieces_platform_client_sdk/.../release_note_detail_view.dart:40
  • …plus 1 more

Upstream alternative?

Upstream still wraps SuperText in IgnorePointer at text.dart:1296 (we conflicted on this exact line during the merge and had to re-assert PR #1). Upstream has not introduced a gesturesEnabled toggle or pointer-passthrough mode. No drop-in replacement.

Verdict

KEEP. Risk to track: upstream will keep re-introducing IgnorePointer on every dev release; expect a conflict on this exact line every 1-2 syncs.

Actionable improvement

• (a) Write a Dart integration test in pieces_flutter_client_sdk that asserts Listener.onPointerDown fires on a paragraph tap — catches future accidental reverts at PR time.
• (b) Open an upstream issue requesting a pointerEventsBehavior: PointerEventsBehavior opt-in on TextComponent.

PR #3 — ContentTapExclusion + isTapAllowedAtDocumentPosition

What it does

1. Marker widget ContentTapExclusion(child: …) so a subtree can opt out of SuperReader tap interception.
2. Helper isTapAllowedAtDocumentPosition({Document, DocumentPosition?}) that returns false when the position has an inline placeholder. Mouse + iOS + Android touch interactors call this automatically.

Where it's exercised

• 1 explicit wrap at frontend/pieces_platform_client_sdk/.../inline/inline_url_builder.dart:137 — the URL chip widget wraps its MouseRegion + ClickableText.onTap (line 164 → URLService.launchURL / file-path opener) in ContentTapExclusion(...) so taps on URL chips open the URL instead of placing a caret.
• Auto-applied via the helper for any AttributedText.placeholders entry; this means all 5 inline builders (inline_url_builder, inline_code_builder, inline_latex_builder, inline_citation_metadata_builder, inline_block) get free tap-exclusion, even though only inline_url_builder wraps explicitly.

Upstream alternative?

Upstream has nothing equivalent. Closest evolution is upstream #2994 (popover toolbar on tablets), but that addresses selection toolbars, not pointer-arena resolution at inline-widget hit-test sites — different problem entirely.

Verdict

KEEP. Strongest candidate for upstream contribution — design is clean (SingleChildRenderObjectWidget marker + interactor predicate), cross-platform (mouse/iOS/Android already wired), and there's an existing 105-line markdown doc at super_editor/lib/super_editor/inline_placeholder_tap_exclusion.md from bb68f4c6.

Actionable improvement

Open an upstream PR. Total diff is ~75 lines of code + 105-line doc — exactly the contribution shape upstream typically accepts.

PR #4 — scrollingEnabled flag

What it does

Adds scrollingEnabled field to SuperReader, DocumentScaffold, DocumentScrollable, and the read-only mouse/iOS/Android touch interactors. When false, disables user scroll AND auto-scroll on selection-drag.

Where it's exercised

• frontend/pieces_platform_client_sdk/.../pieces_markdown_reader.dart:39, 119, 711 exposes scrollEnabled and forwards it.
• 5 hard scrollEnabled: false call sites, all in Copilot:
  • frontend/pieces_copilot_module/.../system_message_renderer.dart:332
  • system_message_renderer.dart:507
  • system_message_renderer.dart:597
  • system_message_renderer.dart:1340
  • frontend/pieces_copilot_module/.../copilot_message_summary_content.dart:214

These are markdown reader embeds inside an outer scrollable (the chat list); without this flag, nested scroll would trap user wheel/trackpad input — a critical desktop UX concern.

Upstream alternative?

Upstream now has SuperMessage (intrinsic-sized, no internal scroll) for chat. Partial replacement is feasible for chat-shaped content (copilot_message_summary_content.dart could plausibly migrate), but not full: the two system_message_renderer.dart sites embed full markdown (images, tables, mermaid, latex) which is more than SuperMessage's chat-message-shaped intrinsic. The scrollingEnabled flag is more general.

Verdict

KEEP, upstream-contribute candidate #2. Pure additive flag — easiest upstream sell.

Actionable improvement

• Open an upstream PR for scrollingEnabled (low risk to upstream).
• Separately, evaluate whether the 5 scrollEnabled: false sites should migrate to SuperMessage post-merge for any chat-shaped ones. That's a follow-up refactor in pieces_copilot_module, not a fork concern.

PR #5 — displayLatexAttribution

What it does

One constant: const displayLatexAttribution = NamedAttribution('display-latex');. 2 lines of code.

Where it's exercised

6 sites, all in pieces_platform_client_sdk:

• widgets/super_editor/components/feather_latex_component.dart:8, 20 — editable LaTeX block
• widgets/super_editor/components/display_latex_component_builder.dart:10, 33 — display-mode LaTeX block
• utils/markdown_to_pdf/markdown_to_pdf_converter_io.dart:476 — LaTeX → PDF
• widgets/super_editor/infrastructure/document_text_to_markdown.dart:529 — LaTeX → markdown
• widgets/super_editor/infrastructure/markdown_deserialization_with_language.dart:387 — Markdown → LaTeX node

This is the whole math-rendering pipeline for Copilot.

Upstream alternative?

Upstream has no native LaTeX/math support. We could move this constant outside super_editor (e.g. into pieces_flutter_client_sdk) and remove the fork delta — but it's a NamedAttribution and they live in super_editor/lib/src/default_editor/attributions.dart for discoverability alongside boldAttribution, codeAttribution, etc. Moving it adds zero value and creates inconsistency.

Verdict

KEEP. 2-line, zero-maintenance change. Lowest-priority of all 5 PRs.

Optional cleanup

Move the constant to pieces_flutter_client_sdk so the fork doesn't carry it at all. Cost: ~5 lines refactor in 6 places. Benefit: 1 less fork delta. Probably not worth doing unless we're already touching all 6 files.

PR #6 — ImageComponent imageBuilder w/ altText + no IgnorePointer wrap

What it does

Three changes to super_editor/lib/src/default_editor/image.dart:

1. Changes imageBuilder signature from (BuildContext, String imageUrl) (positional, upstream) → (BuildContext, {required String imageUrl, String altText}) (named, fork).
2. Removes IgnorePointer wrapper around the SelectableBox (line 247 in fork; upstream still has it at line 233).
3. Adds enableIgnorePointer: selection != null parameter to SelectableBox so the box only ignores pointers when actively selected (so hover works when not selected).

Where it's exercised

Builder consumer at pieces_markdown_reader.dart:721-728:

ImageComponentBuilder(
  imageBuilder: (context, {required String imageUrl, String altText = ''}) {
    return PiecesImageComponent(
      imageUrl: imageUrl,
      altText: altText,
      borderRadius: widget.imageBorderRadius ?? BorderRadius.circular(8),
    );
  },
),

Why pointer-passthrough matters: PiecesImageComponent (components/pieces_image_component.dart:99-105) uses a MouseRegion with onEnter/onExit to track hover state. Line 132 conditionally shows a "Copy URL" button at opacity: _isHovered ? 1.0 : 0.0. That hover state only fires because PR #6 removed IgnorePointer. Reverting this PR would silently break a desktop-only UX feature — no compile error, the button would just never appear.

Upstream alternative?

Upstream now (post our 2026-05 merge):

• ✅ Has altText field on ImageNode (since long ago).
• ✅ Has imageBuilder field on ImageComponent — but positional signature (context, imageUrl) only.
• ❌ Still wraps in IgnorePointer.

Upstream is ~50% there. Two specific deltas remain:

1. Signature mismatch: their (context, imageUrl) vs our (context, {required imageUrl, altText}).
2. IgnorePointer: hard blocker for hover UX.

Verdict

KEEP behavior, contribute API. Pointer-passthrough is load-bearing for desktop. Signature mismatch is cosmetic, but the named-param form (with altText as a typed argument) is better — imageBuilder callers shouldn't have to look up the node to get alt text.

Actionable improvement

• Upstream PR: change imageBuilder signature to (BuildContext, {required String imageUrl, String altText}) — strictly additive.
• Upstream issue (separate): argue for removing the unconditional IgnorePointer in ImageComponent (or making it opt-in via enableIgnorePointer-style param).

Recommended Action Plan

Bucket A — KEEP as-is (high confidence)

• PR feat: remove ignore pointer so gesture detection works #1 (no IgnorePointer in TextComponent): keep, document, add an integration test in pieces_flutter_client_sdk.
• PR [Feat] Display latex attribution added #5 (displayLatexAttribution): keep, low-cost.

Bucket B — KEEP, contribute upstream

• PR Enable tap exclusion for inline widgets in SuperReader #3 (ContentTapExclusion): clean, well-documented, cross-platform — best upstream-contribution shape.
• PR Scroll Enable/Disable Flag #4 (scrollingEnabled): pure additive flag — easiest upstream sell.
• PR [FEAT] Image component updates #6 — API only (named-param imageBuilder w/ altText): clean upstream PR.

Bucket C — Behavior we'll likely have to keep regardless

• PR [FEAT] Image component updates #6 — IgnorePointer removal: even if we contribute the API change, the behavioral change upstream may resist. Plan to keep this delta.
• PR feat: remove ignore pointer so gesture detection works #1: same logic — upstream is unlikely to drop IgnorePointer from TextComponent because it serves their selection model elsewhere. We'll keep re-asserting on each merge.

Suggested upstream PR sequence

1. displayLatexAttribution — trivial 2-line PR, builds rapport.
2. scrollingEnabled flag — additive, isolated.
3. ContentTapExclusion — bigger but clean, has docs.
4. ImageComponentBuilder.imageBuilder named-param signature — strictly more useful than current.
5. Issue (not PR): pointer-passthrough opt-in for TextComponent and ImageComponent.

If upstream merges 1-4, we shrink the permanent fork delta from 5 PRs → 1.5 PRs (the two pointer-passthrough behavioral overrides).

Part 2 — Flutter 3.35-3.40 Compat Investigation (TextInputStyle)

Why this came up

A separate concern surfaced after the merge: super_editor's ime_decoration.dart (post-merge) overrides void updateStyle(TextInputStyle style), but TextInputStyle doesn't exist in our Flutter 3.35.7 SDK — so the file fails to compile. The hypothesis was raised that maybe unset-pub-resolution could route around it by pulling super_editor from a remote source.

Decisive fact-check (Flutter docs + Perplexity)

The local comment originally claimed TextInputStyle was removed in Flutter 3.35+. The opposite is true — it's a future addition, and our Flutter is too old to have it.

Sources verified

1. Flutter docs · Deprecate TextInputConnection.setStyle (last updated 2026-03-03):

   "TextInputConnection.setStyle is deprecated in favor of TextInputConnection.updateStyle… Landed in version: TBD. In stable release: Not yet."

2. Flutter API docs · TextInputConnection (current stable): lists setStyle(...) only. No updateStyle method, no TextInputStyle class.

3. Upstream super_editor PR #2950 description (verbatim, Feb 13, 2026):

   "In [Web] Fix IME and selection by syncing more text styles flutter/flutter#180436 a breaking change was introduced… the framework introduced TextInputConnection.updateStyle(TextInputStyle style) and deprecated the old method… This change fixes the breakage by adding an implementation for the new method."

4. Local Flutter SDK check at /Users/.../flutter-sdk-3.35.5/packages/flutter/lib/src/services/text_input.dart:

   • setStyle({...}) exists at lines 1658, 2429, 2555.
   • updateStyle(TextInputStyle) and class TextInputStyle — do not exist.

The actual situation

Upstream's b422c326 was authored 24 hours after the Flutter master breaking change landed, before stable Flutter caught up. As a result, super_editor 0.3.0-dev.51 has been broken on stable Flutter for ~3 months.

Would unset-pub-resolution route around it?

No. I read automations/pub_resolution.dart (1194 lines). It:

1. Adds/removes resolution: workspace from workspace member pubspecs.
2. On unset, injects standalone dependency_overrides for test, test_api, web_socket_channel, analyzer, dart_style, freezed_annotation, material_symbols_icons — none of which point at super_editor.
3. Strips/restores dependency_overrides de-duplication and build.yaml envied-path blocks.

It does not switch the source of super_editor from local-path to a git tag. The same physical files in frontend/super_editor/super_editor/ get used either way. Hypothesis disproved.

Resolution: 3-line forward-compat patch (commit 73e427c7)

Drop upstream's 3-line override and replace with a documented comment block that records:

• Upstream commit SHA (b422c326)
• Originating Flutter PR (flutter/flutter#180436)
• Link to the breaking-change doc
• Three restore criteria (Flutter stable ships the API → our pin moves past that release → revert)

Why we can't migrate forward instead

TextInputConnectionDecorator implements TextInputConnection. In Flutter 3.35.7:

• TextInputConnection has no updateStyle to override.
• TextInputStyle is not a defined type.
• Dart conditional imports switch entire libraries but cannot synthesize class members based on host SDK version — there's no escape hatch for an interface override that selectively exists.

Why dropping the override is a no-op functionally

setStyle(...) immediately above is still in the framework and still in the parent interface — covers the entire surviving public API on stable Flutter. The framework can't call a method that doesn't exist on the parent.

Tracked

Added as fork divergence #6 in FORK_CHANGES.md, marked TEMPORARY. The Pre-Merge Checklist now flags ime_decoration.dart so future syncs catch upstream re-introducing the override on each merge.

Combined Summary

What the fork delta looks like after this PR + compat patch

Permanent fork delta if all upstream contributions land: 1.5 PRs (#1 + behavioral half of #6).
Permanent fork delta if no upstream contributions land: 5 PRs.
Temporary on top: the updateStyle compat patch, until Flutter upgrade.

Combined recommended next steps

1. Land this PR (already includes both the merge and the compat patch).
2. File 4 upstream PRs in priority order: displayLatexAttribution → scrollingEnabled → ContentTapExclusion → imageBuilder named-param signature.
3. File 1 upstream issue for pointer-passthrough opt-in (TextComponent + ImageComponent).
4. Add integration test in pieces_flutter_client_sdk for the Listener.onPointerDown contract on PiecesMarkdownReader, so an accidental future revert of PR feat: remove ignore pointer so gesture detection works #1's IgnorePointer removal is caught at PR time.
5. When Flutter pin upgrades past TextInputStyle-shipping stable: revert the compat patch (commit 73e427c7).

This is the complete fork hygiene picture as of the 2026-05 sync.