commit 33400a7aba557bc57c3cbde2830326a30c3bf41d
author Aaron Leventhal <aleventhal@chromium.org> Wed Jan 18 17:18:48 2023
committer Chromium LUCI CQ <chromium-scoped@luci-project-accounts.iam.gserviceaccount.com> Wed Jan 18 17:18:48 2023
tree c7129194e86cfb9b6966d585dacf5e69609d101b
parent 0b1b112a4aa3cbfb52fa2784481c14e2d62d39b1

Revert "Reland: [A11y] Stable ids for AXObjects with DOM nodes"

This reverts commit 01466963a42b3bd5fdbe5bc5ac17bff6304b058d.

Reason for revert:
- Performance regressions in crbug.com/1408341
- TAST test inputs.VirtualKeyboardAccent still fails sometimes

Original change's description:
> Reland: [A11y] Stable ids for AXObjects with DOM nodes
>
> Reland of CL:4027071. The revert was caused because of the breaking
> of a TAST test. Basically, some code in AXTreeSerializer that deleted subtrees of old child ids is not triggered as often, because if the DOM node doesn't change, the child ID doesn't. That's a good thing usually, but if the node is reattached with new style, we need to make sure its children are reconsidered. The fix is that in the specific case where
> a node is reattached, call ChildrenChanged() instead of just marking it dirty -- in AXObjectCacheImpl::UpdateCacheAfterNodeIsAttached().
>
> ----
>
> Use the DOM node for the AXID for any AXObject with a DOM node. Other AXObjects will still generate their AXID, but in their own numerical namespace. For now, any AXID < 0 will be a generated ID.
>
> Role changes and alerts required some new browser-side code, because rather than the object being destroyed and created with a new id, the ID now stays the same.
>
> In some cases, test results have improved, e.g. rather than an object being destroyed and created with a different ID, it retains the same ID and a change event is fired for it.
>
> Benefits:
> - Removes over 100 lines of implementation code
> - Use less memory: removes ids_in_use_ (set) and node_object_mapping_ (map).
> - Reduce map lookups when an AXObject is removed -- based on the ID value, we can determine whether it has a DOM node and if it doesn't, we already know it cannot be in some of the maps
> - Make it easier for screen readers to maintain the user's point of regard within content, even when the layout or role changes. Users sometimes lose their place, as the screen reader may move the user to the top. A hope is that this will reduce the occurrences of that happening. It may even be possible for screen readers to keep their place in tabs where the AX tree is released from memory and later regenerated, which could open up possibilities to using less memory.
> - Enable the follow-ups listed below.
>
> Follow-up work:
> - Refactor content-visibility: auto change handling in a11y (crbug.com/1380449).
> - See if the dom_node_id field in AXNodeData is needed anymore now that any id >0 is the same as the dom node id.
> - Look at simplifying or removing reparenting computations in AXTreeSerializer. Maybe we can get rid of expensive AnyDescendantIsReparented().
> - Merge GetAXID() and GetExistingAXID() -- see TODO.
>
> Bug: none
> Change-Id: If6af6cb7e0330e8ab28d7a00bcf5bdfe10fbc9aa
> Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/4131180
> Reviewed-by: David Tseng <dtseng@chromium.org>
> Commit-Queue: Aaron Leventhal <aleventhal@chromium.org>
> Auto-Submit: Aaron Leventhal <aleventhal@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#1093648}

Bug: none
Change-Id: I93adace41a3407118eb49fc71e94b15d747985b3
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/4178414
Auto-Submit: Aaron Leventhal <aleventhal@chromium.org>
Owners-Override: Austin Sullivan <asully@chromium.org>
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Reviewed-by: Aaron Leventhal <aleventhal@chromium.org>
Commit-Queue: Aaron Leventhal <aleventhal@chromium.org>
Cr-Commit-Position: refs/heads/main@{#1093932}