In bug 1925635, I moved the post processing after deleting ranges in
`EditorBase::DeleteRangesWithTransaction` into the `HTMLEditor`'s override.
At this time, I makes it uses `HTMLEditUtils::IsEmptyNode` to check whether
the `Text` and its containers become empty. Therefore, if `Ctrl`+`Backspace`
deletes the only word in the `Text` which starts with a collapsible white-space
causes deleting the `Text` after deleting all visible characters in the `Text`.
Therefore, this makes check it with `Text::TextDataLength` too before searching
the most distant ancestor inline element which becomes empty.
However, it may cause leading invisible white-spaces of the only word visible
because `nsFrameSelection` computes the word range strictly in the `Text`.
Therefore, this patch also makes `HTMLEditor::DeleteRangesWithTransaction` and
`AutoDeleteRangesHandler::ComputeRangesToDeleteRangesWithTransaction` extend
each range to delete to include surrounding invisible white-spaces too.
Therefore, this patch adds the new method to `AutoClonedRangeArray`.
Then, I hit 4 existing bugs with new test failures.
One is `HTMLEditUtils::LineRequiresPaddingLineBreakToBeVisible`. It checks
whether the candidate point to insert a `<br>` is followed by a block boundary
first. Then, it checks whether the candidate point follows a collapsible
white-space or a block boundary. However, it uses
`WSRunScanner::ScanPreviousVisibleNodeOrBlockBoundary()` which ignores invisible
white-spaces. Therefore, it will ignore the new invisible white-space and
reaches preceding `Text`. Thus, it fails to put a `<br>` and makes the new
invisible white-space "fixed" as invisible.
Therefore, this patch rewrites the check with using
`HTMLEditUtils::GetPreviousLeafContentOrPreviousBlockElement()`.
Next one is, `HTMLEditor::DeleteEmptyInclusiveAncestorInlineElements()`
returns end of deleted node if it's followed by a non-editable node, i.e.,
an element has `contenteditable="false"`. Therefore, its caller inserts a
`<br>` to the end of the container when deleting preceding editable node of a
non-editable node.
Therefore, this patch removes the editable state check.
Next, `AutoBlockElementsJoiner::HandleDeleteNonCollapsedRange` may put a padding
`<br>` after the moved line, but it does not assume that the moved line does not
ends with a block boundary. This causes failing #46 and #48 tests in
`text_bug772796.html`. E.g., when pressing `Delete` in
`<div>foo[]<div><span style="white-space:pre"><div>bar</div>baz</span>`, we move
the second `<div>` as a child of the parent `<span>` to end of the first `<div>`
like `<div>foo<span style="white-space:pre"><div>bar</div></span></div>...`.
Without the change, it starts to put unnecessary `<br>` after ` the `<span>`
because of the bug fix in `HTMLEditUtils::LineRequiresPaddingLineBreakToBeVisible`
above. This result is completely odd from the users point of view (looks like
just move caret), but we should avoid to put the unnecessary `<br>`.
Finally, we'll fail an assertion when putting caret at the last of
`AutoBlockElementsJoiner::DeleteContentInRange` because it forgets to flush
the tracking range before using it. This appeared by the changes above.
Therefore, this patch fixes this bug too.
Differential Revision: https://phabricator.services.mozilla.com/D240703
8e23c15e1b