fix: allow backspace to move to next run

[palmer-cl]

This is related to: #2175

• If we try to limit the cursor from moving to another run, we cannot move from runs that contain a single character.
• In the doc listed https://linear.app/superdocworkspace/issue/SD-1942/bug-tracked-changes-deleting-character-deletes-whole-paragraph-chrome, periods are their own run at some points in the doc, meaning that if we backspace next to a period, without this fix, we can never get to the next character.
• Chrome handles this correctly without this fix, but Firefox does not.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d31ddfd06c

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[caio-pizzol]

@palmer-cl makes sense to widen the scan past single-char runs, but removing the bound entirely can delete into a previous paragraph outside of TC mode. left two inline comments — main suggestion is using paragraph start as the bound instead of no bound.

[caio-pizzol]

without a lower bound, this scan can walk past paragraph boundaries and delete a character from the wrong paragraph. that only gets caught by normalizeReplaceStepSingleCharDelete in track-changes mode — in normal editing it goes through silently.

instead of removing minPos entirely, scope it to the current paragraph:

const paragraphStart = $pos.start($pos.depth);
const deleteRange = findPreviousTextDeleteRange(state.doc, $pos.pos, paragraphStart);

that still lets you cross sibling runs (fixing the single-char run issue) without leaving the paragraph.

[palmer-cl]

I did try this as a solution:

here were the results:

[palmer-cl]

@caio-pizzol what are your thoughts?

[caio-pizzol]

the reason $pos.start($pos.depth) didn't work is that $pos.depth points to different levels depending on cursor position:

• between runs (case 1: $pos.nodeBefore is a run) → $pos.depth is the paragraph depth → $pos.start($pos.depth) = paragraph start ✓
• at the start of a run (case 2: $pos.pos === $pos.start()) → $pos.depth is the run's depth → $pos.start($pos.depth) = run content start ✗

so in case 2 the bound collapses back to the run, which is why backspace still couldn't cross into the previous run in your video.

we already solve this exact problem in backspaceAcrossRuns.js:32 — just need to compute paraDepth the same way:

const paragraphDepth = $pos.parent.type === runType ? $pos.depth - 1 : $pos.depth;
const paragraphStart = $pos.start(paragraphDepth);
const deleteRange = findPreviousTextDeleteRange(state.doc, $pos.pos, paragraphStart);

that gives paragraph-scoped bounds for both cursor positions — crosses sibling runs but not paragraph boundaries.

check: #2225

[caio-pizzol]

this test covers the cross-paragraph case — with a paragraph-scoped bound it should still pass. worth keeping.

[palmer-cl]

Closing this PR this is no longer an issue in Firefox post merging main, it looks like findPreviousTextDeleteRange was updated.