You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Greg Hudson <gh...@MIT.EDU> on 2004/06/25 14:42:01 UTC
Re: [Issue 1525] Use new "entity ID" notion instead of "committed
rev"
On Fri, 2004-06-25 at 09:02, kfogel@collab.net wrote:
> brane@tigris.org writes:
> > Another note about terminology:
> >
> > Object or Entity ID is usually used to refer to what we call node IDs -- a
> > single identifier that logically groups related versions of the same object.
>
> Note that in issue #1525, we are using "Entity ID" as a synonym for
> "node revision ID".
Uh, not quite. Issue #1525 proposes that entity IDs be isomorphic to
node revision IDs, not synonymous with them. (Specifically, we use a
change count instead of the transaction ID.)
> In other words, an Entity ID doesn't refer to a
> logically related group of objects, but rather to *one* object (which
> may happen to be reachable from different paths/revisions in the svn
> filesystem).
Correct.
brane wrote:
>> We think we can generate this thing using the filesystem's NodeId, CopyId,
>> and (get this) the PredecessorCount (from the skip-delta code!).
> So you propose to count the predecessors every time. How boring. Why not just
> take the (single logical immediate) predecessor's "change count" (version
> number!) and increment it. I suppose the obvious solutions are too complicated,
> what :-)
No, the obvious solutions are already done. We already store a
predecessor count field in the node revision skel.
> Another 2.0 change is that branch IDs should be unique within node IDs
I don't think you said this like you meant to say it. Branch IDs are
already unique within node IDs; if they weren't, we would lose horribly
when someone committed a change to the same file on two different
branches and wound up with two node revisions with the same node
revision ID.
Perhaps you meant to say that copy IDs should be subsidiary to node IDs,
instead of having an independent cross-node-ID existence like they do in
the current code. That's how I would have done it in the first place,
but (a) perhaps there's some reason for the conservative copy ID
allocation in the current code, and (b) I don't understand how making
copy IDs subsidiary to node IDs makes it easier to implement renames.
This question can wait until the promised 2.0 metamodel document,
though.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [Issue 1525] Use new "entity ID" notion instead of "committed rev"
Posted by kf...@collab.net.
Greg Hudson <gh...@MIT.EDU> writes:
> Uh, not quite. Issue #1525 proposes that entity IDs be isomorphic to
> node revision IDs, not synonymous with them. (Specifically, we use a
> change count instead of the transaction ID.)
Sure, I was just using "synonym" as a synonym for "isomorphic".
(It's not, of course, so really I was using "synonym" isomorphically
to "isomorphic".)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [Issue 1525] Use new "entity ID" notion instead of "committed
rev"
Posted by Greg Hudson <gh...@MIT.EDU>.
On Mon, 2004-06-28 at 01:52, Branko Čibej wrote:
> The point here is that the current branch ID generation scheme is a)
> compliated, and b) has subtle, non-obvious side effects, but doesn't
> have /any/ benefits (that I can see) over the simpler numbering within
> node ID.
It minimizes copies table entries, and its semantics may be important
for the FS history code. If you're considering an alternate schema with
copy IDs subsidiary to node IDs, make sure to work out how the history
functions will work.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [Issue 1525] Use new "entity ID" notion instead of "committed
rev"
Posted by Branko Čibej <br...@xbc.nu>.
C. Michael Pilato wrote:
>>In the current scheme, you can't simply relink a moved (directory)
>>node-rev into another directory, because the copy-id generation
>>algorithm will use the ID from another branch to generate copy-ids for
>>future versions of subsidiary files, thus possibly generating
>>duplicate node-rev-ids. Ooof.
>>
>>
>
>Is *that* the "monstrous blunder"?!
>
>
Yes, it is. But don't worry, I was all for it at the time, remember? I
even went so far as to review and approve Bill Tutt's design. Well, we
all make mistakes.
>Oh.
>
>I can see the mountainous pile of features blocked by this design
>decision. All the functionality put on hold while we sort out the
>grievous error we've made. Time lost. Potential users, turned away
>at the door for lack of Goodness.
>
>(You can stop me at any time.)
>
The point here is that the current branch ID generation scheme is a)
compliated, and b) has subtle, non-obvious side effects, but doesn't
have /any/ benefits (that I can see) over the simpler numbering within
node ID.
-- Brane
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [Issue 1525] Use new "entity ID" notion instead of "committed rev"
Posted by "C. Michael Pilato" <cm...@collab.net>.
Branko Čibej <br...@xbc.nu> writes:
> >No, the obvious solutions are already done. We already store a
> >predecessor count field in the node revision skel.
> >
> Ah, good. I misunderstood Mike's original post, then.
Why yes, yes you did. :-)
> The idea was to get a meaningful branch number, that is, all node-revs
> on the same branch would have the same branch ID (even if they didn't
> belong to the same node!). Of course that turned out to be a monstrous
> blunder.
/me waits for the unsubstantiated claims about the blunder of the
current branch Id design.
> In the current scheme, you can't simply relink a moved (directory)
> node-rev into another directory, because the copy-id generation
> algorithm will use the ID from another branch to generate copy-ids for
> future versions of subsidiary files, thus possibly generating
> duplicate node-rev-ids. Ooof.
Is *that* the "monstrous blunder"?!
Oh.
I can see the mountainous pile of features blocked by this design
decision. All the functionality put on hold while we sort out the
grievous error we've made. Time lost. Potential users, turned away
at the door for lack of Goodness.
(You can stop me at any time.)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: [Issue 1525] Use new "entity ID" notion instead of "committed
rev"
Posted by Branko Čibej <br...@xbc.nu>.
Greg Hudson wrote:
>On Fri, 2004-06-25 at 09:02, kfogel@collab.net wrote:
>
>
>>brane@tigris.org writes:
>>
>>
>
>>So you propose to count the predecessors every time. How boring. Why not just
>>take the (single logical immediate) predecessor's "change count" (version
>>number!) and increment it. I suppose the obvious solutions are too complicated,
>>what :-)
>>
>>
>
>No, the obvious solutions are already done. We already store a
>predecessor count field in the node revision skel.
>
>
Ah, good. I misunderstood Mike's original post, then.
>>Another 2.0 change is that branch IDs should be unique within node IDs
>>
>>
>
>I don't think you said this like you meant to say it. Branch IDs are
>already unique within node IDs; if they weren't, we would lose horribly
>when someone committed a change to the same file on two different
>branches and wound up with two node revisions with the same node
>revision ID.
>
>Perhaps you meant to say that copy IDs should be subsidiary to node IDs,
>instead of having an independent cross-node-ID existence like they do in
>the current code.
>
Yes, that's what I meant.
>That's how I would have done it in the first place,
>but (a) perhaps there's some reason for the conservative copy ID
>allocation in the current code,
>
The idea was to get a meaningful branch number, that is, all node-revs
on the same branch would have the same branch ID (even if they didn't
belong to the same node!). Of course that turned out to be a monstrous
blunder.
>and (b) I don't understand how making
>copy IDs subsidiary to node IDs makes it easier to implement renames.
>
>
In the current scheme, you can't simply relink a moved (directory)
node-rev into another directory, because the copy-id generation
algorithm will use the ID from another branch to generate copy-ids for
future versions of subsidiary files, thus possibly generating duplicate
node-rev-ids. Ooof.
/me waits for his brain to unscramble
>This question can wait until the promised 2.0 metamodel document,
>though.
>
>
Oh yes.
-- Brane
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org