Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [alf-dev] RE: ALF Source Code MangementVocabulary Meeting+1-303-928-3232id 6053141# Wednesday 10:00AM PDT - meeting minutes

For an integrator, what you get back when you ask for changes between to a workspace and a past incarnation thereof is already pretty arbitrary between tools already. I think we'll want to layout a pretty flexible format and just take what we can get from the various tools. The change log will look different based on the tool used.

That's ok because it represents what actual change happened in the system. You don't want CVS to pretend that a delete followed by an add was a rename.

-- Eric

Richard Title wrote:

You may be right. Certainly for an individual command like "rename foo fum"
or "move src/foo bar/foo", it's not too hard to see how to translate that
into equivalent directory operations, if that's what ClearCase needs to do
internally. A more interesting case is a command like "show me all the
differences between branch B1 and branch B2". If ClearCase thinks of
rename's and move's as directory differences, does it have to translate
those into the equivalent file-name/file-parent differences?
Then again, who cares about ClearCase? [I didn't actually say that... :-) ]

Cheers,

Richard

-----Original Message-----
From: alf-dev-bounces@xxxxxxxxxxx [mailto:alf-dev-bounces@xxxxxxxxxxx] On
Behalf Of Eric Minick
Sent: Wednesday, May 03, 2006 4:49 PM
To: ALF Developer Mailing List
Subject: Re: [alf-dev] RE: ALF Source Code MangementVocabulary
Meeting+1-303-928-3232id 6053141# Wednesday 10:00AM PDT - meeting minutes

Richard,

I may just be showing my ignorance of scm systems, but I'm not sure the namespace-handling should be that tough. The integrating tool will be working on a workspace and either it, ALF or both will know the configuration for that workspace. Given a configuration, an ALF command could simply say, "rename /myproject/source/data.c /myproject/source/useful_data.c" perhaps with an element id.

From there it's up to the tool's ALF integration job to worry about what happens. In subversion the operation is on the file; in Clearcase it's on the directory; and in CVS you delete the old file and make a new one. Why should ALF care?

My other thought, and I suppose this is obvious, would be that we may need a command that asks the SCM integration, "What do you support?" Or, "How do you work?". From there we can get into interesting things like meta-data.

But other things are more basic. If I want a fresh copy of Project X, in our use cases one would:
1) Create the workspace
2) Matterialize the workspace

That's fine in a lot of systems, but in CVS creating the workspace (cvs checkout) implicitly populates it as well. Step 2 could be skipped. ALF will need to know that.

A sub use case of diffing a workspace would be to just ask, "are there any changes to this workspace?". Going back to CVS, getting a full revision listing can be an expensive operation. Some versions (and only some) of CVS have a seperate command for "are there changes?" that would save a workflow time if there are not. Right now, integrating tools either ignore this or solve it with a check box. Other tools are in a similar boat. Last time I used Dimensions (feel free to correct me) from the command line it was pretty easy to get a list of the last time each file had changed. So figuring out if anything had changed since some date was easy. But to get the details of those changes was considerably more expensive.

I suspect that as we go, we'll hit more situations where some tools have combined two concepts that ALF might otherwise assume are atomic. One option is to have them report their behavior. The other is to make ALF use cases at a larger grain. If instead of:
1) Create a workspace
2) Materialize a workspace

We had:
1) Create and populate a workspace.
2) Update a workspace (re-materialize)

ALF would need to know much less about the SCMs it was integrating with, but they would be more likely to operate in a sub-optimal way. Both approaches work fine.

I've rambled quite long enough. Sorry about that.

-- Eric


Richard Title wrote:

...


We ran out of time at the end to discuss some of these, but Brian Carroll included managing metadata as a possible use case. I could see Subversion


users potentially wanting this feature as metadata can be used for a variety of purposes. In Subversion metadata is associated with a file or folder and can be essentially anything including binary content, such as thumbnail image for a JPEG.
Agreed. I had in mind use case 7 (create new version of Element) as
potentially covering changes in the data (i.e. the file-content), and/or
changes in the metadata (a.k.a. properties, a.k.a. attributes) of the
Element. Metadata could include filename, parent directory, modification
time, owner, permissions, file-type, thumbnail-image, etc. See Concepts,
item 2 (File & Directory concepts).

The part that can get thorny is namespace-handling. Is renaming a element
an
operation on the element itself (as AccuRev does), or on the directory
containing it (as ClearCase does)? Is moving a file to a new directory an
operation on the element being moved (AccuRev), or on the parent
directories
involved (ClearCase)? What's Subversion's model?
Richard

_______________________________________________
alf-dev mailing list
alf-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/alf-dev



_______________________________________________
alf-dev mailing list
alf-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/alf-dev

_______________________________________________
alf-dev mailing list
alf-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/alf-dev



Back to the top