The difference here seems to be that previously a trailing space was excluded from p-wrapping, so that when
the serializer looked at the constrains to determine if a newline was necessary, it suppressed it. The new
behaviour seems preferable, since this is a block. The rt classification is broken since the two aren't really
semantically different.

'dewiki/Klingspor (Unternehmen)',

T89628 might be related

T196931 is a case where making the encapsulation marker transparent to p-wrapping is hindering instead of helping. We probably want to migrate them *in* for some scenarios.

In T200098#4445959, @Arlolra wrote:

T196931 is a case where making the encapsulation marker transparent to p-wrapping is hindering instead of helping. We probably want to migrate them *in* for some scenarios.

Unless I am misunderstanding, you are basically suggesting DOM templating by opening/closing paragraphs on template boundaries. Doing that just in Parsoid will lead to different rendering in PHP parser and Parsoid.

If so, I think you should treat that as a different issue from this ticket.

No, that's not what I'm suggesting. Look at openPTag in the ParagraphWrapper.js, it's careful about expanding template ranges unnecessarily. The decision to unilaterally treat template start metas as transparent in the dom based pass is suspect. It should depend on whether that template gets closed within the paragraph.

In T200098#4446007, @Arlolra wrote:

No, that's not what I'm suggesting. Look at openPTag in the ParagraphWrapper.js, it's careful about expanding template ranges unnecessarily. The decision to unilaterally treat template start metas as transparent in the dom based pass is suspect. It should depend on whether that template gets closed within the paragraph.

As far as PHP parser semantics are concerned, these start/end meta tags are indeed transparent to p-wrapping.

But, in Parsoid land, we add additional logic to push these off subtree edges ... so in this case, what you are recommending is that we should start/end paragraphs on these meta tags (don't start on an end tag; don't end on a start tag). That make sense to me. That also continues to fit the core model of what this phab task is about.

Yes, the semantics as is are correct, but the outcome is less than desirable.

As in the linked task, there're cases where including the start tag in the paragraph will result in more precise ranges.