It's occurred to me that the extra height might be intentional, i.e., to make room for the highlighting that should be present but is missing. If that's the case, then this could be closed as Declined and the problem with new content not being highlighted can be merged to one about diffing tables (or this one re-titled and re-purposed for that).

I did another test. Just click on the visual edit button on the article:
*https://en.wikipedia.org/wiki/List_of_countries_by_infant_and_under-five_mortality_rates?veaction=edit

Before any editing is done the empty header cell above the row numbers has had its height increased. It is easy to tell because the row number column rows now no longer align with the table rows to the right. Due to the increase in the header cell height.

I rarely use the visual editor, so when I started noticing all this misalignment before and after editing I started the Mediawiki thread. I now realize that the VisualDiff is going to be difficult no matter what. It has long been difficult to decipher many wikitext diffs. So I understand how it could be even more difficult to show additions and removals with a visual editor. Especially in tables.

So I would be happy if there was no height increases in cells before clicking on publish or review. It is supposed to be a WYSIWYG visual editor.

This problem with row height being increased would not be a problem if there were not a separate row-number table adjacent to the main table.

There would be no misalignment of the rows if the column of row numbers was integral to, and part of, the main table.

So if T42618 was resolved then this problem here would not be a real problem either. In fact, the row height increase would be helpful (along with possible highlighting) as a way to show the Visual Diff. So I suggest finishing this enhancement:

T42618 : jquery.tablesorter: Add support for a "fixed" column of row numbers.

This isn't reproducible on that page any more after the table in it was changed, and when trying to reproduce with old revisions, I always get an error that "It took too long to compute your changes". A minimal test case would be appreciated.

This issue might actually be fixed by the fix for T295252.