Thanks for the fix, it seems to do the trick.

I have some comments on the new indentation implementation, though:

1. In getLineIndex and getParentLineIndex:

while ( index-- ) {

Since index is provided by the function caller, I would guard this more safely with index-- <= 0.

2. In getParentLineIndex:

if ( formatsAtIndex.length === startFormats.length - 1 ) {

Just from reading, this seems to encompass any kind of rich-text format, not just ul / ol. Is there a risk that this will fail (i.e. that we have a false-positive) for certain rich-text values if there happens to be some other format (like strong) whose edge sits right at LINE_SEPARATOR?

Since index is provided by the function caller, I would guard this more safely with index-- <= 0.

Could you elaborate on why this is safer?

Just from reading, this seems to encompass any kind of rich-text format, not just ul / ol. Is there a risk that this will fail (i.e. that we have a false-positive) for certain rich-text values if there happens to be some other format (like strong) whose edge sits right at LINE_SEPARATOR?

I'm not sure I'm following. Could you illustrate with an example?

Just from reading, this seems to encompass any kind of rich-text format, not just ul / ol. Is there a risk that this will fail (i.e. that we have a false-positive) for certain rich-text values if there happens to be some other format (like strong) whose edge sits right at LINE_SEPARATOR?

Ah I think I get what you mean. Formats at line separators should only be line formats like ul / ol. It's a good point to raise though. While the normal formats are ignored when building the DOM or HTML tree, they do get applied through applyFormat. We should prevent application there.

I'm not sure I'm following. Could you illustrate with an example?

Simply put, if something somewhere (plain human error, or some other unexpected evaluation) ends up calling getLineIndex( value, -1 ), the VM will break.

While the normal formats are ignored when building the DOM or HTML tree, they do get applied through applyFormat. We should prevent application there.

Great, thanks!

Simply put, if something somewhere (plain human error, or some other unexpected evaluation) ends up calling getLineIndex( value, -1 ), the VM will break.

Ah, the selection can't be negative, so I didn't add logic for it. Should I account for it?