phases: update the phase set as we go during retract boundary
Apparently iterating over the `changed_revs` dictionary is very expensive.
On mozilla-try-2019-02-18, a perf::unbundle call with a 10 000 changesets
bundle gives give use the following timing.
e57d4b868a3e: 4.6 seconds
ac1c75188440: 102.5 seconds
prev-changeset: 30.0 seconds
this-changeset: 4.6 seconds
So, the performance regression is gone.
Once again: thanks to marvelous Python!
$ cat >> $HGRCPATH << EOF
> [storage]
> dirstate-v2.slow-path=allow
> [format]
> use-dirstate-v2=no
> EOF
Set up a v1 repo
$ hg init repo
$ cd repo
$ echo a > a
$ touch file-with-somewhat-long-name-to-make-dirstate-v1-bigger-than-v2
$ hg commit -Aqm a
$ hg debugrequires | grep dirstate
[1]
$ ls -1 .hg/dirstate*
.hg/dirstate
Copy v1 dirstate
$ cp .hg/dirstate $TESTTMP/dirstate-v1-backup
Upgrade it to v2
$ hg debugupgraderepo -q --config format.use-dirstate-v2=1 --run | grep -E 'added:|removed:'
added: dirstate-v2
$ hg debugrequires | grep dirstate
dirstate-v2
$ ls -1 .hg/dirstate*
.hg/dirstate
.hg/dirstate.* (glob)
Manually reset to dirstate v1 to simulate an incomplete dirstate-v2 upgrade
$ rm .hg/dirstate*
$ cp $TESTTMP/dirstate-v1-backup .hg/dirstate
There should be no errors, but a v2 dirstate should be written back to disk
$ hg st
$ ls -1 .hg/dirstate*
.hg/dirstate
.hg/dirstate.* (glob)
Corrupt the dirstate to see how the errors show up to the user
$ echo "I ate your data" > .hg/dirstate
$ hg st
abort: working directory state appears damaged! (no-rhg !)
(falling back to dirstate-v1 from v2 also failed) (no-rhg !)
abort: Too little data for dirstate: 16 bytes. (rhg !)
[255]