hbisect: do not assume that min(good) is an ancestor of min(bad)

The included test used to report "inconsistent state", which is
incorrect. While this situation cannot occur when the user sticks to
the suggested bisect sequence. However, adding more consistent
good/bad information to the bisect state should be tolerated as well.

mercurial/hbisect.py

@@ -35,8 +35,9 b' def bisect(changelog, state):'
         # build visit array
         ancestors = [None] * (len(changelog) + 1) # an extra for [-1]
 
-        # set nodes descended from goodrev
-        ancestors[goodrev] = []
+        # set nodes descended from goodrevs
+        for rev in goodrevs:
+            ancestors[rev] = []
         for rev in xrange(goodrev + 1, len(changelog)):
             for prev in clparents(rev):
                 if ancestors[prev] == []:

tests/test-bisect2.t

@@ -431,3 +431,22 b' end at merge: 17 bad, 11 good (but 9 is '
   date:        Thu Jan 01 00:00:09 1970 +0000
   summary:     9
   
+
+user adds irrelevant but consistent information (here: -g 2) to bisect state
+
+  $ hg bisect -r
+  $ hg bisect -b 13
+  $ hg bisect -g 8
+  Testing changeset 11:82ca6f06eccd (3 changesets remaining, ~1 tests)
+  2 files updated, 0 files merged, 0 files removed, 0 files unresolved
+  $ hg bisect -g 2
+  Testing changeset 11:82ca6f06eccd (3 changesets remaining, ~1 tests)
+  0 files updated, 0 files merged, 0 files removed, 0 files unresolved
+  $ hg bisect -b
+  The first bad revision is:
+  changeset:   11:82ca6f06eccd
+  parent:      8:dab8161ac8fc
+  user:        test
+  date:        Thu Jan 01 00:00:11 1970 +0000
+  summary:     11
+