Very strange map edit - please help investigate. Maybe rollback?

Here’s a changeset I was working on last night: Please disregard the comment as it reflects only a very small portion of work that went into the changeset. I kept making changes and forgot to jump out of the changeset.

Among other things I did was edit a relation which I had created shortly before that for a suburb named Woodbury. Let’s not get into whether Woodbury is technically a suburb or what a suburb is. It seems to fit the bill and it was suggested to me in another thread by SK53, so that’s that.

In the initial version, it had a polygon member with role=boundary and a node tagged place=suburb with role=label. It worked fine, but I decided to try something else. I removed the polygon and instead used the four streets serving as natural (and official) boundaries of the suburb as role=boundary members. It seems to work well enough and I tested the relation through the OSM relation layer.

Today when I looked at the rendered map, the four streets that made up the boundary of the suburb became moved by so many meters. It is as if someone selected all members and dragged them off their location. Moreover, when I go into the OSM data, it tells me that the roles of the four streets in the relation have changed from “boundary” as they were in v3 to “” in v4. And it tells me that this happened in my changeset and no one else has touched this data.

Now, I can conceive that I highlighted the four streets and accidentally nudged them off their location, but I don’t believe I changed the roles as well.

Can anyone tell me who, how, when and why made these two changes (move and role change) to my four streets?

I don’t know if this change can be rolled back (or if rollbacks are done on changeset level). I would not want the whole changeset rolled back. It is easy enough for me to manually move the four streets back into their place if that’s the most efficient fix.

UPDATE: 8 hours later, I could not look at it anymore and moved the streets back in position. I have not fixed the roles in the relation and still would like any insight into what happened.

To track down reasons for road shifts, look at the changesets and check the history of some key nodes that you would not have expected to change location. I’m not sure if Potlatch will warn you when shifting roads, but JOSM frequently catches accidental road shifts and warns me before moving them to be sure I really meant to move them.

The history just says that I moved them in the changeset that I listed above. Fine. It hardly explains how the member roles in the relation went from “boundary” to “”.

I am going to guess that Potlatch2 glitched up and am not going to worry about it unless it happens again.

I was hoping that someone with better understanding (or deeper access) than me could look at the history in a different way and tell me more about what happened. Someone wrote once that on OSM all changes are well document and can be rolled back.

Here’s one of the few nodes that jumped:

Says I did it. I say I did it at Tue, 12 Apr 2011 06:38:44

Here’s the relation:

The member ways were added at Tue, 12 Apr 2011 06:05:39 as “boundary” and changed to “” at Tue, 12 Apr 2011 06:38:44 (so, same as the jumping streets).

Even if I highlighted and accidentally dragged the four streets, which I don’t think I did, I know I did not change the roles to “”.

Could you have accidentally added a second copy of each member to the relation, without any role (so each member now had two different roles, “boundary” and blank)?

Anything’s possible at this point. I know I said I’m not going to worry about it and I will try to live by these words.