Improved PR diffs

Hello,

We’ve all experienced the issue during review of PR diffs that contain
spurious changes. I believe I’ve tracked down why and fixed it. Or
improved it.

In brief: I was trying to be clever: attempting to ensure that the diffs
were only the ones relevant since the branch you last rebased from. I
did that because I previously persuaded myself that some of the diff
problems we were seeing were caused by divergence between where the PR
diffed from and where the PR was currently.

That probably improves the diffs if you *haven’t* done a rebase off
master recently.

Buuuuuttttt…the actual diffs are off the current consensus drafts in
HTML, and those always have all of the merged PRs.

So I’ve taken that cleverness out. If you push changes to your PR
without first rebasing off the current master, you’re going to be
introducing spurious diffs. But if you rebase off the current master
before you push, the diffs should be only the ones relevant to your PR.

That’s my theory until we see evidence to the contrary!

                                        Be seeing you,
                                          norm

--
Norm Tovey-Walsh
Saxonica