perf: run the gc before each run
The python garbage collector is a large source of performance troubles, we run
it right before the timed section to reduce the change for the gc to add noise
to the benchmark.
Checking how hg behaves when one side of a pull/push doesn't support
some capability (because it's running an older hg version, usually).
$ hg init repo1
$ cd repo1
$ echo a > a; hg add -q a; hg commit -q -m a
$ hg bookmark a
$ hg clone -q . ../repo2
$ cd ../repo2
$ touch $TESTTMP/disable-lookup.py
$ disable_cap() {
> rm -f $TESTTMP/disable-lookup.pyc # pyc caching is buggy
> cat <<EOF > $TESTTMP/disable-lookup.py
> from mercurial import extensions, wireprotov1server
> def wcapabilities(orig, *args, **kwargs):
> cap = orig(*args, **kwargs)
> cap.remove(b'$1')
> return cap
> extensions.wrapfunction(wireprotov1server, '_capabilities', wcapabilities)
> EOF
> }
$ cat >> ../repo1/.hg/hgrc <<EOF
> [extensions]
> disable-lookup = $TESTTMP/disable-lookup.py
> EOF
$ hg pull ssh://user@dummy/repo1 -r tip -B a
pulling from ssh://user@dummy/repo1
no changes found
$ disable_cap lookup
$ hg pull ssh://user@dummy/repo1 -r tip -B a
pulling from ssh://user@dummy/repo1
abort: other repository doesn't support revision lookup, so a rev cannot be specified.
[255]
$ disable_cap pushkey
$ hg pull ssh://user@dummy/repo1 -r tip -B a
pulling from ssh://user@dummy/repo1
abort: remote bookmark a not found!
[10]