SVN timing

[Kevin Puetz]

Nikolas Zimmermann wrote:

> On Thursday 24 March 2005 00:28, Richard Moore wrote:
>> On Wed, 23 Mar 2005 23:32:12 +0100, Dirk Mueller <mueller at kde.org> wrote:
>> > loosing "checking out by date" ability. no way to find a commit by
>> > date,
>> > no way to reliably diff  the repository by date.
>>
>> This is a MAJOR loss from my point of view. I often use this facility.
>> :-(
> I agree, I also use this quite often to completely diff a project against
> a version, say two days ago.

Well, I'll put two points out.
1) with the global revision number, most of the cases where you previously
had to use dates go away - you can do svn log on the whole folder, and then
diff against the revision representing the point you wanted to compare to.
Before you couldn't do that (versions were per-file), so dates were the
only choice.

2) And anyway, dates should work fine. The discussion about them not working
had to do with incrementally converting modules (since this would mean that
"old" dates will get dropped into "new" revision numbers, and confuse it).
AFAIK, this the main reason the decision was made to do it all at once
instead. I'm not aware of any other reason why they won't work given this
decision... did something else come up?

>> > Ah, also we'll experience a factor 2-3 slowdown of all repository
>> > operations.
>>
>> That doesn't sound good either.
> Agreed.

I can't speak to this in the large scale; I certainly would rank the speed
of svn (which makes status&diff - aka "what have I got sitting here?" -
into local operations) as better than cvs on the operations that I use
*all* the time. blame is the only thing noticeably worse (and it's not
*that* much worse in my experience, though svn does have some scaling
issues on this command). However, the CVS repo (at work) that I converted
was only about 600M - not near the size of KDE. Where did that 2-3x number
come from?

> It sounds like we really shouldn't move to SVN.
> The only thing I'd like to ask: why hasn't this mentioned before?!
> (In case I missed someone pointing out this stuff before, forgive me..)
> 
> All the fuzz about a new version control system, which doesn't really
> have _that_ much benefits for us it seems *grml* *grml* *grml*
> 
> Bye
>  Bye
>   Niko