Skip to content

Touch drag: remaining physical-device, OS-cancel, IME and multitouch verification #412

Open

Description

@superlin2017

Verification status after v1.29.13

The bounded swipe/hold and touch-focus fixes shipped in #466: v1.29.13, release commit c91cb720. The same-spot displacement and empty-Undo claims were not reproduced on the investigated baseline; drop geometry and history rollback were not changed. #411 tap-to-edit remains separate product policy.

Native XCTest touch in iPhone17e/iOS26.5 Simulator Safari passed quick swipe, both no-op/Redo paths, deliberate 13→10→14 reorder, and persistence across Safari restart/reload at b97ff300. The final candidate 5a0ceeb8 retained the touch handler, bullet CSS, and drop callback; surrounding main changes were reconciled and checked with 515 app tests, 39 real-Worker tests, and green hosted CI/CLI. This is not an exact-final-head native run or native before/after proof. Evidence and limits.

Remaining verification, keeping this issue open:

  • Physical-device touch behavior.
  • Explicit OS touch cancellation/app-switch, beyond observed scroll pointercancel.
  • Native IME and multitouch interactions.
  • Other iOS versions and a native baseline comparison.

The separate physical Undo obligation remains open. Production promotion succeeded; two public smoke-check passes, one minute apart and separate from the deployment workflow, confirmed both sites at v1.29.13, homepages/entry scripts and auth configuration responding, and unauthenticated outline reads denied. No production outline data was changed.

Original report (retained for history)

What happened?​

Three problems in the same gesture model. The bullet dot is at once the zoom
target, the drag handle, and part of the left margin where a thumb starts a
swipe. Code involved: src/components/use-drag-reorder.ts,
src/components/OutlineEditor.tsx, src/styles.css.

1. A swipe that starts on the dot cannot scroll — it reorders instead

.bullet sets touch-action: none, so when a touch begins on the dot the
browser is never allowed to scroll. The dot sits in the left margin, exactly
where a left-handed thumb starts a swipe, so "scroll the page" regularly becomes
"drag this row". The drag arms after 5px of movement (THRESHOLD), which is
inside normal finger jitter, and consumeClick() then swallows the click — the
node has been moved, the zoom never happens, and nothing tells the user why.

2. A drag released without moving still moves the node

beginDrag removes the grabbed row from the drag's row model, so everything
below slides up under a stationary finger; the projection then reads the gap
under that finger as "one row down". onUp submits pending unconditionally,
so picking a row up and putting it back down in the same spot drops it one
position lower.

There is a second half to this: onMove calls capture() before moveNode().
moveNode correctly returns false for a same-spot drop, but the undo entry
has already been pushed — so the user gets an undo step that restores nothing,
and pressing undo appears to do nothing at all.

3. A reorder puts the moved row into edit mode, raising the on-screen keyboard

After a successful move, onMove sets both pendingFlash (the landing tint)
and pendingFocus (caret back in the moved row). The caret is right for a
keyboard-first desktop app; on a phone it means every reorder ends with the
on-screen keyboard sliding up out of what was a pure gesture, and the viewport
jumps. The tint already shows where the row landed, so the caret adds nothing
on touch.

What did you expect to happen?​

  • A swipe that starts on the dot should scroll the page.
  • A tap on the dot should zoom in.
  • Picking a row up and releasing it where it started should change nothing, and
    should not add an undo step.
  • Only a deliberate drag should reorder.
  • After a reorder, the row should be tinted at its new position but should not
    enter edit mode on a touch device.

Steps to reproduce

Problem 1 (touch):

  1. On a phone, start a swipe with your thumb on the left edge of the screen, so
    the touch begins on a bullet dot.
  2. Swipe up or down.
  3. The page does not scroll; the row is dragged and reordered instead.

Problem 2 (touch or mouse):

  1. Press a bullet dot and move just past the drag threshold (a few pixels), so
    the row is picked up.
  2. Release without moving any further.
  3. The node lands one position lower even though the finger barely moved.

Problem 3 (touch):

  1. On a phone, long-press a bullet dot and drag the row to a new position.
  2. Release.
  3. The row moves and flashes, but the on-screen keyboard opens and the row is
    left in edit mode.

  • Page: https://app.dotflowy.com/
  • Browser: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/155.0.0.0 Safari/537.36
  • Screen: 1920x919
  • When: 2026-10-17T09:13:46.860Z

Possible direction (not a prescription — just what seemed to work in a local
patch):

  • Let the dot pan vertically (touch-action: pan-y) and arm the drag only after
    a short hold, taking the pan back with a non-passive touchmove.
  • On release, skip the move when the pointer ended where it started, or when the
    resolved gap equals the node's current parent + predecessor.
  • Gate the post-move caret on pointer type (matchMedia("(pointer: coarse)"))
    and keep the flash unconditional — FocusPass handles focus and flash
    separately, so the landing tint still works with no focus.

Activity

  1. pullfrog commented on Oct 17, 2026

    @pullfrog
    Contributor
  2. changed the title [-]Touch: swiping on the bullet dot reorders instead of scrolling, and a same-spot drop still moves the node[/-] [+]Touch: dot swipes reorder instead of scrolling, same-spot drops still move the node, and a reorder raises the keyboard[/+] on Oct 17, 2026
  3. cameronapak commented on Oct 26, 2026

    @cameronapak
    Owner

    Investigated unchanged main at 9d1c60e, including #411, ADRs 0019/0029/0069, merged drag/history changes, and reference-only PR #311.

    • Dot swipe: reproduced in Chromium touch emulation. A 120px swipe from the dot left scrollY at 0 and entered drag instead of scrolling.
    • Same-spot drop / empty undo: not reproduced in the tested mouse cases (6px pickup/release and drag away/return). Order and an existing redo action survive. Current drag geometry excludes the source as a target without collapsing its occupied space, and the editor already calls the capture rollback when moveNode returns false. No geometry or history rewrite is proposed.
    • Post-drop focus: reproduced in Chromium touch emulation. A successful reorder focuses the moved contentEditable. This proves focus, not an actual iPhone keyboard opening.
    • Mobile: Tapping a node auto-focuses the editor and opens the keyboard, making it hard to just view/browse #411: tapping row text/dead space to edit is the explicit ADR 0029 behavior and remains a separate product choice.

    Draft #466 limits the fix to native bullet scrolling until a deliberate touch hold and suppressing the post-drop editor-focus request for touch. Focused browser regressions pass; physical iPhone before/after verification is queued with the existing device owner. Keeping this issue open until native evidence and delivery are complete. No native-device pass is claimed.

  4. cameronapak commented on Oct 26, 2026

    @cameronapak
    Owner

    Candidate b97ff30 now has positive iPhone17e / iOS26.5 Simulator evidence in normal Safari using native XCTest trusted touch, not mouse or JS-dispatched input. I verified the delivered archive hash, node projections, history logs, reload hashes, and recording frames.

    • Immediate dot swipe: observed 117px over 227ms, scrollY 337→701; all 43 node projections unchanged.
    • Hold 500ms + 6px, and 100px away/exact return: unchanged data; populated Redo preserved, Undo remained disabled.
    • Deliberate Touch10 drop: correct 13→10→14 order; only Touch10/11/14 predecessor links changed, with parents/text unchanged. Exact projection survives Safari termination/reopen/reload.
    • Instrumented cases ended with BODY active and native keyboard count 0. Pristine-shell swipe/reorder controls also passed using external API, native accessibility, and visual evidence; no visible keyboard/caret/callout or remaining horizontal drag indicator.

    Limits: baseline 9d1c60e was not executed in the Simulator, so this is candidate verification, not native before/after proof. Explicit OS cancellation/app-switch, physical-device, IME, multitouch, and other-iOS coverage remain unproved. Stale pristine collector state/trace files are excluded. Physical Undo verification is a separate obligation.

    Keeping #412 narrowed/open through delivery and the remaining unproved cases. #411 browse/tap policy is unchanged. Draft #466 is updated from delivered v1.29.9 main at head 209d964; 8 targeted browser cases pass, fresh exact-head CI is pending. The simulator result remains attributed to b97ff30, not the updated head. Release slot remains after #245, #237, and #191 with coordinator clearance.

  5. changed the title [-]Touch: dot swipes reorder instead of scrolling, same-spot drops still move the node, and a reorder raises the keyboard[/-] [+]Touch drag: remaining physical-device, OS-cancel, IME and multitouch verification[/+] on Oct 27, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions