feat(input): block-editing pipeline foundation by wildseansy · Pull Request #460 · software-mansion/react-native-enriched-markdown

wildseansy · 2026-06-29T02:40:04Z

What

First PR in the block-editing series requested in #455 (closed in favor of this rebuild). This adds the generic block-editing pipeline foundation for the editable input, mirroring the existing inline-style pipeline (StyleType → FormattingStore → StyleHandler → InputFormatter → serializer/parser). No concrete block type is registered yet — with no handler the editor behaves exactly as today (plain paragraphs, all inline styles unaffected). A follow-up PR plugs the first handler (headings, H1–H6) in without touching the orchestrator.

Abstractions (both platforms)

Inline (existing)	Block (this PR)
`ENRMInputStyleType` / `StyleType`	`ENRMInputBlockType` / `BlockType` (Paragraph default)
`ENRMStyleHandler` / `StyleHandler`	`ENRMBlockHandler` / `BlockHandler` — paragraph attributes, markdown line prefix (parser recognition stays in each parser's central map, mirroring the inline pipeline)
`ENRMFormattingStore` / `FormattingStore`	`ENRMBlockStore` / `BlockStore` — generic, line-scoped block ranges + `adjustForEdit`
formatter style-handler registry	formatter block-handler registry (empty here) + `applyBlockRanges`
serializer inline delimiters	block handlers contribute line prefixes (no central switch)
parser inline span mapping	iOS: real md4c `enter_block`/`leave_block` (replaces the no-op TODO); Android: block ranges from the shared AST (C++ untouched)

Notes addressing the #455 review

No monolithic orchestrator logic — blocks plug in via handlers, like bold did for inline.
iOS parser uses md4c enter_block/leave_block (the existing TODO is removed), not a post-pass.
Android reuses the shared MarkdownASTNode the readonly renderer already emits — MD4CParser.cpp is unchanged, so readonly parsing is unaffected.
No silent serializer fallback — the block-aware serializer asserts the line-count invariant instead of silently dropping blocks.

Testing

Both platforms build green (example app, iOS sim + Android emulator). Behavior is inert (byte-for-byte unchanged) with no block handler registered.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EhuNE6PjRGjroxobhdd72q

Add the generic block-editing abstractions for the input component, mirroring the inline-style pipeline (StyleType→FormattingStore→StyleHandler →InputFormatter→serializer/parser). No concrete block type is registered yet, so behavior is unchanged (plain paragraphs, inline styles intact); a later change plugs a heading handler in without touching the orchestrator. - ENRMInputBlockType enum + block paragraph attribute keys - ENRMBlockRange model (type + range + generic level payload) - ENRMBlockHandler protocol: paragraph attributes, markdown line prefix, md4c block matching — sufficient for headings and list items - ENRMBlockStore: generic line-scoped block range store with edit adjustment - InputFormatter: block-handler registry (empty) + applyBlockRanges - Serializer: block-aware overload; asserts the line-count invariant instead of silently dropping blocks - Parser: real md4c enter_block/leave_block via a kSupportedBlocks table (paragraph-only today), replacing the no-op TODO - Orchestrator: wires a block store through apply/edit/import/export, inert until a handler is registered Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EhuNE6PjRGjroxobhdd72q

Android mirror of the iOS block-editing foundation: the generic block abstractions for the input component, paralleling the inline-style pipeline (StyleType→FormattingStore→StyleHandler→InputFormatter→serializer/parser). No concrete block type is registered, so behavior is unchanged; a later change plugs a heading handler in without touching the orchestrator. - BlockType enum (PARAGRAPH default) + BlockRange model (type/start/end/level) - BlockHandler interface: createSpans, spanClasses, markdownLinePrefix, matchesNodeType — sufficient for headings and list items - BlockStore: generic line-scoped block range store with edit adjustment (reuses FormattingStore's edit-overlap logic) - InputFormatter: block-handler registry (empty) + applyBlockFormatting - MarkdownSerializer: block-aware overload; asserts the line-count invariant instead of silently dropping blocks - InputParser: ParseResult.blockRanges from the shared AST (paragraph-only today; the C++ parser is unchanged so readonly rendering is unaffected) - Orchestrator + event emitter: wire a block store through apply/edit/ import/export, inert until a handler is registered Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EhuNE6PjRGjroxobhdd72q

wildseansy · 2026-06-29T02:46:24Z

To show how this foundation is meant to be used, here's the first handler (headings, H1–H6) plugging into it as a reference PR, stacked on this branch so the diff is only the heading delta:

wildseansy#1

It demonstrates the intended extension pattern — a new block type is added by implementing the BlockHandler protocol (paragraph styling + markdown line prefix + parser-block matching) and registering it, with no orchestrator changes. Per-level heading sizes are configured via markdownStyle.h1..h6, mirroring the readonly renderer. Verified on both platforms (clean editor text, per-level sizing, single-marker round-trip). Lists would follow the same pattern.

wildseansy · 2026-06-29T05:35:01Z

@hryhoriiK97 - base pipeline ready for review. See wildseansy#1 for how it's used for headings, and wildseansy#2 for the headings fully integrated into the test app

hryhoriiK97

@wildseansy thanks for the PR! I'll try to take a look this week if possible. We have a lot on our plate right now, so I can't make any promises.

- drop the unused handler-matching API (matchesNodeType / matchesMd4cBlockType / allBlockHandlers); parser recognition stays in each parser's central map, mirroring the inline pipeline (nodeTypeToStyleType / kSupportedSpans) - derive iOS inline + block ranges from a single md4c run (was two full parses per import) - iOS applyBlockRanges: strip the previous pass's attributes (tracked via ENRMBlockTypeAttributeName) before re-applying, and seed the paragraph style from existing attributes instead of replacing it - align the serializer line-count invariant contract: log + fall back to inline-only output on both platforms instead of crashing release Android via check() - clamp paragraphBoundsForRange unconditionally so out-of-bounds input can't reach paragraphRangeForRange: (iOS) - document the store's non-overlap invariant and insert-at-range-start semantics on both platforms Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

hryhoriiK97 · 2026-07-03T09:48:23Z

+    return index
+  }
+
+  private enum class EditOverlap {


adjustForEdit, EditOverlap, and classifyOverlap are copy-pasted from FormattingStore. Any future bug fix to overlap logic would need to be applied in both places. Worth extracting into a shared utility that both stores can delegate to.

Extracted into a shared unit on both platforms in 5019270: RangeEditAdjustment (Kotlin, operates on a MutableRangeBounds interface both range models implement) and ENRMRangeEditAdjustment (iOS). Both stores now delegate; the duplicated EditOverlap/classifyOverlap copies are gone.

hryhoriiK97 · 2026-07-03T09:48:23Z

+ * @property level generic integer payload, 0 by default. Headings use it for the
+ *   H-level (1-6); list items will use it for nesting depth.
+ */
+data class BlockRange(


This is a data class with mutable var fields, which means equals()/hashCode() depend on mutable state. If a BlockRange ever ends up in a set or as a map key, mutations will silently break lookups. Same pattern as FormattingRange so not new to this PR, but worth keeping in mind.

Converted both BlockRange and FormattingRange to plain classes in 5019270 — bounds mutate as edits shift ranges, so identity semantics are the safe default. No call site relied on structural equality (removeRange receives the store's own instance).

hryhoriiK97 · 2026-07-03T09:48:23Z

+    NSRange range = rangeValue.rangeValue;
+    [textStorage removeAttribute:ENRMBlockTypeAttributeName range:range];
+    [textStorage removeAttribute:ENRMBlockLevelAttributeName range:range];
+    [textStorage removeAttribute:NSParagraphStyleAttributeName range:range];


The reset pass strips NSParagraphStyleAttributeName from previously-claimed ranges, so the "seed from existing attributes" on re-apply will always get the default paragraph style rather than whatever was there before. In practice this is probably fine since applyWritingDirection runs after and handlers set absolute values - just something to be aware of.

Good catch — the comment overpromised. Reworded in 5019270: seeding only matters for newly claimed paragraphs; re-claimed ones read the default after the reset pass, which is fine since handlers set absolute values and applyWritingDirection re-derives direction afterwards.

hryhoriiK97 · 2026-07-03T10:11:20Z

  private fun serializeToMarkdown(): String {
    val plainText = view.text?.toString() ?: ""
-    return MarkdownSerializer.serialize(plainText, view.allFormattingRangesForSerialization())
+    // Each block resolves its markdown line prefix through its registered handler.


Nit: this comment narrates what the code does — the function call below is self-explanatory. Consider removing.

Removed in 5019270.

hryhoriiK97 · 2026-07-03T10:11:20Z

+      return inlineMarkdown
+    }
+
+    // Plain-text character offset at the start of each line.


Nit: the variable name lineStartOffsets already says this. Consider removing.

Removed in 5019270 (also from the iOS twin).

hryhoriiK97 · 2026-07-03T10:11:20Z

+    var runningOffset = 0
+    for (i in plainLines.indices) {
+      lineStartOffsets[i] = runningOffset
+      runningOffset += plainLines[i].length + 1 // +1 for the '\n' separator


Nit: obvious from context, consider removing.

Removed in 5019270.

hryhoriiK97 · 2026-07-03T10:11:20Z

+      for (lineIndex in plainLines.indices) {
+        val lineStart = lineStartOffsets[lineIndex]
+        val lineEnd = lineStart + plainLines[lineIndex].length
+        // A block claims a line if their ranges intersect (block ranges are


Nit: the if condition is clear enough on its own. Consider removing.

Removed in 5019270 (also from the iOS twin).

hryhoriiK97 · 2026-07-03T10:11:20Z

+                     ranges:(NSArray<ENRMFormattingRange *> *)ranges
+                blockRanges:(NSArray<ENRMBlockRange *> *)blockRanges
+{
+  // The provider runs synchronously inside the serializer call, so capturing


Nit: over-explaining a trivial local capture. Consider removing.

Removed in 5019270.

- extract the duplicated shift/clip edit-adjustment logic into one shared unit per platform (RangeEditAdjustment / ENRMRangeEditAdjustment); FormattingStore and BlockStore now delegate to it on both platforms - convert FormattingRange and BlockRange from data classes to plain classes: their bounds mutate as edits shift ranges, so value-based equals/hashCode would silently break set/map lookups - correct the applyBlockRanges seeding comment: re-claimed paragraphs read the default style after the reset pass; seeding matters only for newly claimed paragraphs - drop comments that narrate what the adjacent code already says (InputEventEmitter, MarkdownSerializer on both platforms, EnrichedMarkdownTextInput) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

wildseansy · 2026-07-03T14:26:07Z

@hryhoriiK97 thanks for the review — everything is addressed, replies on each thread. Summary of the updates:

5019270 — your review:

Extracted the duplicated adjustForEdit/EditOverlap/classifyOverlap logic into one shared unit per platform: RangeEditAdjustment (Kotlin, via a small MutableRangeBounds interface both range models implement) and ENRMRangeEditAdjustment (iOS). FormattingStore and BlockStore now delegate on both platforms — net −91 lines.
Converted BlockRange and FormattingRange from data classes to plain classes so equals/hashCode don't depend on mutable bounds. Verified no call site relied on structural equality.
Fixed the misleading "seed from existing attributes" comment in applyBlockRanges.
Removed all the narrating comments you flagged (plus their identical twins in the iOS serializer).

c3d7992 — earlier self-review pass, in case you're re-reviewing the whole diff:

Dropped the unused handler-matching API (matchesNodeType / matchesMd4cBlockType / allBlockHandlers); parser recognition lives in each parser's central map, mirroring the inline pipeline. (The extension-pattern note in my earlier comment above predates this — a new block type now adds one map entry + one handler.)
iOS parses markdown once per import instead of twice (block ranges come from the same md4c run as inline spans).
iOS block pass now strips its previous attributes before re-applying, so removed blocks can't leave stale paragraph styling.
Serializer line-count invariant no longer crashes release Android — both platforms log + fall back to inline-only output.

Verified: ktlint + compileDebugKotlin green, iOS builds clean via xcodebuild, and the maestro smoke suite passes on functional assertions (screenshot-threshold diffs on my machine are iOS-runtime font drift vs the 26.3 baselines, unrelated to this diff).

hryhoriiK97

LGTM! Thank you!

@wildseansy are you planning to open the heading handler PR (from wildseansy#1) as a follow-up?

wildseansy · 2026-07-03T16:09:39Z

@hryhoriiK97 - yep. I'll re-assign base to software:main. Thanks for the review!

wildseansy · 2026-07-03T16:35:15Z

@hryhoriiK97 - heading blocks submitted: #497. Subsequent PR here that adds into the toolbar: wildseansy#2

…into feat/list-blocks Rebases the bullet-list work onto the upstream-merged pipeline. The v2 branch carries main (0.7.3) plus the squashed foundation (software-mansion#460) and heading handler, including the maintainer refactors the list branch predated — resolutions: - adopt the shared RangeEditAdjustment / ENRMRangeEditAdjustment and move the anchored-block (heading / bullet) persistence out of the old per-store switch: stores delegate to the shared adjustment and layer the anchor rules around it (zero-length anchors kept at / shifted past / dropped with the edit; a block deleted exactly to its end collapses to an anchor at the edit location) - generalize normalizeToLineBounds to keep empty-line anchors of persisting block types (BlockType.ANCHORED / ENRMBlockTypePersistsWhenEmpty); it supersedes normalizeAnchoredRangesToLines (Android) and clipHeadingBlocksToFirstParagraph (iOS) - iOS dropStaleEmptyBlockAnchors becomes pruneOrphanedBlockAnchors, matching Android's prune: any persisting block no longer at a line start reverts its merged line to a plain paragraph; this also fixes interior empty-line anchors being dropped (the old check required paragraph.length == 0, never true for a line with a terminator) and covers the paste path - adopt the central-map parser recognition (drop matchesNodeType / matchesMd4cBlockType from heading + list handlers, drop allBlockHandlers), the single md4c run on iOS, and the serializer log-and-fall-back invariant (composed with the list ZWSP stripping; also fixes the Kotlin ZWSP constant being an escaped literal instead of the U+200B character) - iOS setBlockType: stores content-only bounds (trims the line terminator), matching the parser convention - keep main-side additions (copyToClipboard command, PasteboardUtils, context-menu changes) alongside the list layer Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(input): heading blocks (H1-H6) on the block pipeline First concrete block type on the #460 foundation. Adds HeadingBlockHandler / ENRMHeadingBlockHandler (paragraph styling + markdown line prefix), maps md4c/AST heading nodes in each parser's central block map, exposes per-level h1-h6 style props mirroring the readonly renderer's markdownStyle, and re-normalizes block ranges to whole-line bounds after edits. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * refactor(input): address review — scope iOS re-formatting to edited lines, cache heading fonts - iOS now re-applies inline + block attributes only on the line(s) touched by an edit (parity with Android's applyFormattingScopedToEdit). Scoping the block pass alone would break: the inline pass stamps baseFont over its range, so both passes share the same line-expanded scope. Layout invalidation shrinks to the scope as well. - Cache heading fonts per level in ENRMInputFormatterStyle, invalidated with the trait font cache on base-font change and per level on config change. - Replace ENRM_APPLY_HEADING_PROPS macro with a template helper, matching the file's existing applyInputStyleProps pattern. - Trim/remove narrating and duplicated comments flagged in review. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

wildseansy and others added 2 commits June 28, 2026 18:44

wildseansy mentioned this pull request Jun 29, 2026

feat(input): heading blocks (H1–H6) — first consumer of the block pipeline wildseansy/react-native-enriched-markdown#1

Open

wildseansy marked this pull request as ready for review June 29, 2026 05:35

wildseansy mentioned this pull request Jun 29, 2026

feat(input): heading toolbar, commands & docs — third piece of the block series wildseansy/react-native-enriched-markdown#2

Open

hryhoriiK97 reviewed Jun 29, 2026

View reviewed changes

hryhoriiK97 reviewed Jul 3, 2026

View reviewed changes

hryhoriiK97 approved these changes Jul 3, 2026

View reviewed changes

hryhoriiK97 merged commit 13898e0 into software-mansion:main Jul 3, 2026
9 checks passed

wildseansy mentioned this pull request Jul 3, 2026

feat(input): heading blocks (H1-H6) on the block pipeline #497

Merged

wildseansy mentioned this pull request Jul 4, 2026

feat(input): heading toolbar support — toggleHeading command, active-state, empty-line persistence #499

Open

Uh oh!

Conversation

wildseansy commented Jun 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What

Abstractions (both platforms)

Notes addressing the #455 review

Testing

Uh oh!

wildseansy commented Jun 29, 2026

Uh oh!

wildseansy commented Jun 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hryhoriiK97 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

hryhoriiK97 Jul 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

wildseansy commented Jul 3, 2026

Uh oh!

hryhoriiK97 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

wildseansy commented Jul 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

wildseansy commented Jul 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

wildseansy commented Jun 29, 2026 •

edited

Loading

wildseansy commented Jun 29, 2026 •

edited

Loading

hryhoriiK97 Jul 3, 2026 •

edited

Loading

wildseansy commented Jul 3, 2026 •

edited

Loading

wildseansy commented Jul 3, 2026 •

edited

Loading