Something fishy with branch head

[Greg Ward]

I'm trying some basic workflow operations on a trial conversion from a
fairly large CVS repo (108,000 changesets, 169 branches, 17,000)
files).  Right now I'm trying to ensure that all recent release
branches are merged up to the trunk: i.e. release i merged to release
i+1 merged to release i+2 ... merged to trunk (aka default).

But I'm getting confused by some very basic stuff, and I hope I'm not
seeing a bug.

First, here's the actual sequence of branches that I want to merge:

PACS-3-7-1
PACS-3-7-2
PACS-3-8-1
PACS-3-8-2
default

So my first step is to merge PACS-3-7-1 to PACS-3-7-2.  'hg branches'
claims this branch is active, i.e. has a head:

$ /usr/local/hg-1.3/hg branches -a | head
default                   108707:11b2726ebb69
PACS-3-8-2                108702:7fff910c0944
PACS-3-7-2                108691:1c4d4b7de4ba
PACS-3-8-1                108682:cf76faa059c4
PACS-3-7-1                108665:a860d470aa81
[...many more branches...]

But that's not right: there exists a merge changeset on another branch
whose second parent is 108665:

$ hg glog --template "{rev}:{node|short}   {branches}\n"
[...]
| o | | |  108667:70d522ddc703   PACS-3-8-1-GA
| |\| | |
| | o | |    108666:05338dc3c58e   PACS-3-7-2
| | |\ \ \
| | | o | |  108665:a860d470aa81   PACS-3-7-1
| | | | | |
| +-----o |  108664:6d7eedb13051   PACS-3-8-2
[...]

That is, 108665 is a branch head (it has no children on branch
PACS-3-7-1), but it is not a repo head (it does have children on other
branches).  Alternate (non-graphical) proof:

$ hg log -r108666 --template "{rev}:{node|short}   {branches}\n"
108666:05338dc3c58e   PACS-3-7-2
$ hg parents -r108666 --template "{rev}:{node|short}   {branches}\n"
108658:6b9a8f6326de   PACS-3-7-2
108665:a860d470aa81   PACS-3-7-1

Or, another way:

$ hg heads -q | grep 108665
[no output: 108665 is *not* a head]

So, is 'hg branches' wrong about PACS-3-7-1 being an "active" branch?
Or am I misunderstanding things?

Incidentally, I'm seeing the same result with Mercurial 1.1.1, 1.2.1,
and 1.3.1.  I also get the same result if I delete
.hg/branchheads.cache and re-run 'hg branches'.

Anyways, this confusion leads to other confusion.  For example,
'merge' and 'merge --preview' behave differently:

$ /usr/local/hg-1.3/hg update PACS-3-7-2
0 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ /usr/local/hg-1.3/hg merge --preview PACS-3-7-1
changeset:   108665:a860d470aa81
branch:      PACS-3-7-1
parent:      108657:3b8cad0071af
[...]

but

$ /usr/local/hg-1.3/hg merge PACS-3-7-1
abort: can't merge with ancestor

>From looking at the graph, it appears that 'merge' is right, 'merge
--preview' is wrong, 'heads' is right, and 'branches' is wrong.  Aiee!

Any clue what's going on?  Thanks --

Greg