[p4] Confused by 'p4 integrate' behavior.

[Weintraub, David]

One of the problems with Perforce is that directories aren't full fledge
members of your archive. Instead, they're merely "attributes" that are
attached to each file. Each time Perforce looks at your integration, it
sees no //depot/x/branchOld/snafu, so it assumes that
//depot/x/branchNew/snafu should also be removed to keep the two
branches in sync. There's no way for Perforce to track the idea that
snafu should be in one branch, but not the other since -- unlike file
merges and integrations -- this information isn't in the file itself.

If Perforce truly versioned directories as well as files, integrations
could also be performed on the directories themselves. Perforce could be
made to understand that the directory version on branchNew has a file in
it that isn't on the directory version of branchOld, and Perforce would
stop insisting on deleting snafu on branchNew every time you do your
integration.

As you discovered, the way to handle this issue is to manually
manipulate the branchspec. You also have to do this if you rename the
file on branchNew or moved the file to another directory.

I would personally like true directory versioning -- mainly because I
come from the ClearCase world and understand how it works. However,
directory versioning creates other issues that would complicate
integrations and merges, so I can understand why Perforce chose to avoid
it.

Directory versioning can create some weird side effects that confuses
many developers, and that may be the reason why Perforce chose not to
implement it. I've noticed that Perforce tends to implement things in a
way that Developers find easier to understand (like interfile branching)
rather than ways I think as a CM they "should" be implemented.

-----Original Message-----
From: perforce-user-bounces at perforce.com
[mailto:perforce-user-bounces at perforce.com] On Behalf Of Krzysztof
Kozminski
Sent: Tuesday, October 03, 2006 5:57 AM
To: perforce-user at perforce.com
Subject: [p4] Confused by 'p4 integrate' behavior.

Here's the relevant portion of the history of two files:

% p4 filelog //depot/x/branchNew/snafu
//depot/x/branchNew/snafu
... #1 change 4509 add on 2006/10/03 ...

% p4 filelog //depot/x/branchOld/snafu
//depot/x/branchOld/snafu
... #5 change 4292 delete on 2006/08/31 ...
... #4 change 4285 edit on 2006/08/31 ...
... #3 change 4278 edit on 2006/08/30 ...
... #2 change 4246 edit on 2006/08/25 ...
... #1 change 4241 add on 2006/08/25 ...

Relevant info: //depot/x/branchNew was branched off //depot/x/ branchOld
in change 4506, using this branch specification:

    //depot/x/branchOld/... //depot/x/branchNew/...

and this command:

    p4 integrate -b branchNew -s #have

while portions of branchOld were synced @4500, some portions were @4490,
and the rest was up-to-date (yes, there was a good reason for that). I
am positive that snafu was not present in any revision prior to #5.

Now, the confusing part: the integration attempt:

% p4 integrate -n //depot/x/branchOld/... //depot/x/branchNew/...

attempts to delete snafu from branchNew.

It seems to me that given the sequence of events (branchOld/snafu was
deleted BEFORE branchNew/snafu was added), the integration has no
business deleting branchNew/snafu.

If branchNew/snafu was branched off some revision < 5 of branchOld/
snafu, then yes, deletion would be understandable, but not in the
scenario described above.

Two questions:

1) Is this supposed to be happening that way? What is the rationale
here?

2) How do I tell Perforce to ignore the deletion that happened on
branchOld, keep the stuff I have in branchNew, and not attempt to delete
branchNew/snafu? I am looking here for something like 'p4 resolve -ay'
that would ignore 'their' change (in this case, 'their'  
deletion). While I can hack a branch specification like this:

    //depot/x/branchOld/... //depot/x/branchNew/...
    -//depot/x/branchOld/snafu //depot/x/branchNew/snafu

this looks like too much effort for something that should be doable with
less hassle (I have another 50 files like 'snafu'). Also, if I ever add
branchOld/snafu again, I would like to get a chance to integrate it
without having to remember to remove it from the branch specification...

BTW, both client and server are version 2005.1.

KK
_______________________________________________
perforce-user mailing list  -  perforce-user at perforce.com
http://maillist.perforce.com/mailman/listinfo/perforce-user