Re: [PATCH in progress] Ref-counting for pristine texts

The complete patch (with log message) is attached. Any review would be
appreciated.

> I have tried some timing comparisons and can't even see a difference.

My test WC consisted of 10 identical directories, each containing 1000
different text files, each consisting of a single line "Hello NNN" where
NNN is a number from 000 to 999.

Timings for the patched version (with the ref-count verification code
disabled):

CMD: svn add --quiet wc1/d000/ wc1/d001/ wc1/d002/ wc1/d003/ wc1/d004/
wc1/d005/ wc1/d006/ wc1/d007/ wc1/d008/ wc1/d009/
3.22 real, 1.52 user, 1.71 sys
CMD: svn commit --quiet -m wc1
86.05 real, 62.47 user, 19.80 sys
CMD: svn st --quiet wc1
0.30 real, 0.16 user, 0.10 sys
CMD: svn co --quiet file:///home/julianfoad/tmp/svn/ref1/repo wc2
28.67 real, 15.09 user, 10.30 sys
CMD: svn up --quiet -r0 wc2
5.98 real, 2.78 user, 2.31 sys

Timings for the unpatched trunk code (r1057552):

CMD: svn add --quiet wc1/d000/ wc1/d001/ wc1/d002/ wc1/d003/ wc1/d004/
wc1/d005/ wc1/d006/ wc1/d007/ wc1/d008/ wc1/d009/
3.30 real, 1.33 user, 1.98 sys
CMD: svn commit --quiet -m wc1
117.22 real, 63.88 user, 36.82 sys
CMD: svn st --quiet wc1
0.32 real, 0.14 user, 0.11 sys
CMD: svn co --quiet file:///home/julianfoad/tmp/svn/ref0/repo wc2
29.66 real, 15.08 user, 10.49 sys
CMD: svn up --quiet -r0 wc2
5.99 real, 2.56 user, 2.32 sys

Some of these timings - especially the commit - varied widely across
several runs, both patched and unpatched.

- Julian