Fix long-standing excessive file merges

Since switching to the multihead approach, we've been creating
excessive file-level merges where files are marked as merged with
their ancestors.

This explicitly checks at commit time whether the two parent versions
are linearly related, and if so, reduces the file check-in to a
non-merge. Then the file is compared against the remaining parent,
and, if equal, skips check-in of that file (as it's not changed).

Since we're not checking in all files that were different between
versions, we no longer need to mark so many files for merge. This
removes most of the 'm' state marking as well.

Finally, it is possible to do a tree-level merge with no file-level
changes. This will happen if one user changes file A and another
changes file B. Thus, if we have have two parents, we allow commit to
proceed even if there are no file-level changes.

tests/test-filebranch

@@ -0,0 +1,73 b''
+#!/bin/sh
+
+# This test makes sure that we don't mark a file as merged with its ancestor
+# when we do a merge.
+
+cat <<'EOF' > merge
+#!/bin/sh
+echo merging for `basename $1`
+EOF
+chmod +x merge
+
+echo creating base
+hg init a
+cd a
+echo 1 > foo
+echo 1 > bar
+echo 1 > baz
+echo 1 > quux
+hg add foo bar baz quux
+hg commit -m "base" -d "0 0"
+
+cd ..
+hg clone a b
+
+echo creating branch a
+cd a
+echo 2a > foo
+echo 2a > bar
+hg commit -m "branch a" -d "0 0"
+
+echo creating branch b
+
+cd ..
+cd b
+echo 2b > foo
+echo 2b > baz
+hg commit -m "branch b" -d "0 0"
+
+echo "we shouldn't have anything but n state here"
+hg debugstate | cut -b 1-16,35-
+
+echo merging
+hg pull ../a
+env HGMERGE=../merge hg update -vm --debug
+
+echo 2m > foo
+echo 2b > baz
+echo new > quux
+
+echo "we shouldn't have anything but foo in merge state here"
+hg debugstate | cut -b 1-16,35- | grep "^m"
+
+hg ci -m "merge" -d "0 0"
+
+echo "main: we should have a merge here"
+hg debugindex .hg/00changelog.i
+
+echo "foo: we should have a merge here"
+hg debugindex .hg/data/foo.i
+
+echo "bar: we shouldn't have a merge here"
+hg debugindex .hg/data/bar.i
+
+echo "baz: we shouldn't have a merge here"
+hg debugindex .hg/data/baz.i
+
+echo "quux: we shouldn't have a merge here"
+hg debugindex .hg/data/quux.i
+
+echo "everything should be clean now"
+hg status
+
+hg verify

tests/test-filebranch.out

@@ -0,0 +1,58 b''
+creating base
+creating branch a
+creating branch b
+we shouldn't have anything but n state here
+n 644          2 bar
+n 644          3 baz
+n 644          3 foo
+n 644          2 quux
+merging
+pulling from ../a
+searching for changes
+adding changesets
+adding manifests
+adding file changes
+added 1 changesets with 2 changes to 2 files
+(run 'hg update' to get a working copy)
+merging for foo
+resolving manifests
+ force None allow 1 moddirstate True linear False
+ ancestor a0486579db29 local ef1b4dbe2193 remote 336d8406d617
+ remote bar is newer, get
+ foo versions differ, resolve
+getting bar
+merging foo
+resolving foo
+file foo: other 33d1fb69067a ancestor b8e02f643373
+we shouldn't have anything but foo in merge state here
+m 644          3 foo
+main: we should have a merge here
+   rev    offset  length   base linkrev nodeid       p1           p2
+     0         0      73      0       0 cdca01651b96 000000000000 000000000000
+     1        73      68      1       1 f6718a9cb7f3 cdca01651b96 000000000000
+     2       141      68      2       2 bdd988058d16 cdca01651b96 000000000000
+     3       209      66      3       3 9da9fbd62226 f6718a9cb7f3 bdd988058d16
+foo: we should have a merge here
+   rev    offset  length   base linkrev nodeid       p1           p2
+     0         0       3      0       0 b8e02f643373 000000000000 000000000000
+     1         3       4      1       1 2ffeddde1b65 b8e02f643373 000000000000
+     2         7       4      2       2 33d1fb69067a b8e02f643373 000000000000
+     3        11       4      3       3 aa27919ee430 2ffeddde1b65 33d1fb69067a
+bar: we shouldn't have a merge here
+   rev    offset  length   base linkrev nodeid       p1           p2
+     0         0       3      0       0 b8e02f643373 000000000000 000000000000
+     1         3       4      1       2 33d1fb69067a b8e02f643373 000000000000
+baz: we shouldn't have a merge here
+   rev    offset  length   base linkrev nodeid       p1           p2
+     0         0       3      0       0 b8e02f643373 000000000000 000000000000
+     1         3       4      1       1 2ffeddde1b65 b8e02f643373 000000000000
+quux: we shouldn't have a merge here
+   rev    offset  length   base linkrev nodeid       p1           p2
+     0         0       3      0       0 b8e02f643373 000000000000 000000000000
+     1         3       5      1       3 6128c0f33108 b8e02f643373 000000000000
+everything should be clean now
+checking changesets
+checking manifests
+crosschecking files in changesets and manifests
+checking files
+4 files, 4 changesets, 10 total revisions

mercurial/hg.py

@@ -840,8 +840,30 b' class localrepository:'
                 tm = util.is_exec(self.wjoin(f), mfm.get(f, False))
                 r = self.file(f)
                 mfm[f] = tm
-                mm[f] = r.add(t, {}, tr, linkrev,
-                              m1.get(f, nullid), m2.get(f, nullid))
+
+                fp1 = m1.get(f, nullid)
+                fp2 = m2.get(f, nullid)
+
+                # is the same revision on two branches of a merge?
+                if fp2 == fp1:
+                    fp2 = nullid
+
+                if fp2 != nullid:
+                    # is one parent an ancestor of the other?
+                    fpa = r.ancestor(fp1, fp2)
+                    if fpa == fp1:
+                        fp1, fp2 = fp2, nullid
+                    elif fpa == fp2:
+                        fp2 = nullid
+
+                    # is the file unmodified from the parent?
+                    if t == r.read(fp1):
+                        # record the proper existing parent in manifest
+                        # no need to add a revision
+                        mm[f] = fp1
+                        continue
+
+                mm[f] = r.add(t, {}, tr, linkrev, fp1, fp2)
                 if update_dirstate:
                     self.dirstate.update([f], "n")
             except IOError:
@@ -879,19 +901,20 b' class localrepository:'
             commit = c + a
             remove = d
 
-        if not commit and not remove and not force:
+        p1, p2 = self.dirstate.parents()
+        c1 = self.changelog.read(p1)
+        c2 = self.changelog.read(p2)
+        m1 = self.manifest.read(c1[0])
+        mf1 = self.manifest.readflags(c1[0])
+        m2 = self.manifest.read(c2[0])
+
+        if not commit and not remove and not force and p2 == nullid:
             self.ui.status("nothing changed\n")
             return None
 
         if not self.hook("precommit"):
             return None
 
-        p1, p2 = self.dirstate.parents()
-        c1 = self.changelog.read(p1)
-        c2 = self.changelog.read(p2)
-        m1 = self.manifest.read(c1[0])
-        mf1 = self.manifest.readflags(c1[0])
-        m2 = self.manifest.read(c2[0])
         lock = self.lock()
         tr = self.transaction()
 
@@ -918,6 +941,26 b' class localrepository:'
             r = self.file(f)
             fp1 = m1.get(f, nullid)
             fp2 = m2.get(f, nullid)
+
+            # is the same revision on two branches of a merge?
+            if fp2 == fp1:
+                fp2 = nullid
+
+            if fp2 != nullid:
+                # is one parent an ancestor of the other?
+                fpa = r.ancestor(fp1, fp2)
+                if fpa == fp1:
+                    fp1, fp2 = fp2, nullid
+                elif fpa == fp2:
+                    fp2 = nullid
+
+                # is the file unmodified from the parent?
+                if not meta and t == r.read(fp1):
+                    # record the proper existing parent in manifest
+                    # no need to add a revision
+                    new[f] = fp1
+                    continue
+
             new[f] = r.add(t, meta, tr, linkrev, fp1, fp2)
 
         # update manifest
@@ -1715,12 +1758,7 b' class localrepository:'
                 self.ui.status("(use update -m to merge across branches" +
                                " or -C to lose changes)\n")
                 return 1
-            # we have to remember what files we needed to get/change
-            # because any file that's different from either one of its
-            # parents must be in the changeset
             mode = 'm'
-            if moddirstate:
-                self.dirstate.update(mark.keys(), "m")
 
         if moddirstate:
             self.dirstate.setparents(p1, p2)
@@ -1739,7 +1777,7 b' class localrepository:'
                 self.wfile(f, "w").write(t)
             util.set_exec(self.wjoin(f), mf2[f])
             if moddirstate:
-                self.dirstate.update([f], mode)
+                self.dirstate.update([f], 'n')
 
         # merge the tricky bits
         files = merge.keys()

tests/test-merge6.out

@@ -6,7 +6,7 b' adding file changes'
 added 1 changesets with 1 changes to 1 files
 (run 'hg update' to get a working copy)
 bar should remain deleted.
-6b70e9e451a5a33faad7bbebe627e46b937b7364 644 foo
+f405ac83a5611071d6b54dd5eb26943b1fdc4460 644 foo
 pulling from ../A2
 searching for changes
 adding changesets
@@ -15,4 +15,4 b' adding file changes'
 added 1 changesets with 0 changes to 0 files
 (run 'hg update' to get a working copy)
 bar should remain deleted.
-6b70e9e451a5a33faad7bbebe627e46b937b7364 644 foo
+f9b0e817f6a48de3564c6b2957687c5e7297c5a0 644 foo

tests/test-push-warn.out

@@ -18,4 +18,4 b' searching for changes'
 adding changesets
 adding manifests
 adding file changes
-added 2 changesets with 2 changes to 2 files
+added 2 changesets with 1 changes to 1 files

tests/test-tags.out

@@ -10,6 +10,4 b' acb14030fe0a+ first'
 acb14030fe0a21b60322c440ad2d20cf7685a376+ first
 M a
 c8edf04160c7 tip
-c8edf04160c7+b9154636be93+ tip
-M .hgtags
-M a
+c8edf04160c7+b9154636be93 tip