test-subrepo-recursion.t: re-introduce glob to handle \ slashes on Windows Regression was introduced in 6047947afb6b. Resolved by using the fine line noise '\r (no-eol) (glob) (esc)'.

run-tests: make it possible to combine (esc) with (glob) and (re) This makes it possible to combine the annotations ... if done in the right order.

run-tests: alternative way of handling \r on Windows After f71d60da58fb all \r was stripped from output on Windows, and the places where a \r explicitly was expected it was accepted that it was missing. Ugly hack. Instead we now accept that an extra \r might appear at the end of lines on Windows. That is more to the point and less ugly.

obsolete: flip `obstore.successors` and `obsolete.precursors` People were confused by the fact `obstore.precursors` contained marker allowing to find "precursors" and vice-versa. This changeset changes their meaning to: - precursors[x] -> set(markers on precursors edges of x) - successors[x] -> set(markers on successors edges of x) Some documentation is added to clarify the situation.

obsolete: add example of marker usage in the documentation Recent discussion with Augie Fackler pointed the lack of such example in the documentation.

obsolete: cheap detection of nullid as successors Nullid as successors create multiple issues: - Nullid revnum is -1, confusing algorithm that use revnum unless you add special handling in all of them. - Nullid confuses "divergent" changeset detection and resolution. As you can't add any successors to Nullid without being in even more troubles Fortunately, there is no good reason to use nullid as a successor. The only sensible meaning of "succeed by nullid" is "dropped" and this meaning is already covered by obsolescence marker with empty successors set. However, letting some nullid successors to slip in may cause terrible damage in such algorithm difficult to debug. So I prefer to perform and clear detection of of such pathological changeset. We could be much smarter by cleaning up nullid successors on the fly but it would be much for expensive. As core Mercurial does not create any such changeset, I think it is fine to just abort when suspicious situation is detected. Earlier experimental version created such changesets, so there are some out there. The evolve extension added the necessary logic to clean up its mess.