tests: optimise creating repositories and loading dumps Previously, we'd use svnadmin for creating repositories and loading dumps. That tends to be a bit slow, as it forks a new process and loads the Subversion libraries into it. Instead, we extend our existing Subversion wrappers and load the dumps using the API. This is a noticable speedup. The only downside is that we rely on Subversion and Subvertpy to correctly close all file descriptors; an assumption which hasn't always held in the past. I ran some benchmarks on my relatively slow Mac with $TMPDIR on a ramdisk, and they showed a significant change: I compared ten runs of each with Subvertpy: min: -18.8% (299.1s -> 243.0s) median: -20.0% (307.1s -> 245.6s) ...and three runs of each with SWIG: min: -22.8% (368.7s -> 284.7s) median: -25.7% (384.4s -> 285.5s) (Since the timing measures wall clock time, the minimum time is likely to be the most accurate and useful measurement.)

#!/bin/sh
#
# Generate copies.svndump
#

rm -rf temp
mkdir temp
cd temp
mkdir -p import/trunk/dir
echo a > import/trunk/dir/a

svnadmin create testrepo
svnurl=file://`pwd`/testrepo
svn import import $svnurl -m init

svn co $svnurl project
cd project
svn cp trunk/dir trunk/dir2
echo b >> trunk/dir2/a
svn ci -m 'copy/edit trunk/dir/a'
svn up
svn cp trunk/dir2 trunk/dir3
svn ci -m 'copy dir2 to dir3'
svn rm trunk/dir3/a
svn cp trunk/dir2/a trunk/dir3/a
svn ci -m 'copy and remove'
cd ..

svnadmin dump testrepo > ../copies.svndump