Re: Some svn numbers

[Mike Stump <mrs at apple dot com>]

On Oct 12, 2005, at 5:55 PM, Marcin Dalecki wrote:
> On 2005-10-12, at 04:42, Daniel Berlin wrote:
> > Checkouts will be about 30% slower with svn, just because it has to
> > write more data out to disk because of the working copy
>
> Yes. Indeed. One suggestions comes immediately to my mind. Why  
> don't you provide some kind of COW (Copy on Write)? Or maybe one  
> should call it - checkout a copy on demand?

How does one portably do this?  How does one non-portably do this?  :-)

( Yes, I know, unionfs and the like, but across a reboot it dies. )

In emacs, you set version control, and then just use *.~1~, side  
benefit:

$ cd gcc
$ time mkdiff .  # find . -name \*.~1~ | while read f; do diff -c  
$f ...; done
real    0m1.654s
user    0m0.248s
sys     0m0.946s

versus:

$ time cvs diff .
real    0m7.645s
user    0m0.631s
sys     0m2.082s

4.6x slower.

$ time svn diff .
real    0m16.380s
user    0m1.361s
sys     0m6.291s

9.9x slower.  Wait, I thought svn diff was supposed to be faster?  Hum.