view contrib/hammerclient.py @ 5778:84affb254cdf

evolvecmd: don't update working copy when using in-memory merge

This patch removes the update of the working copy after each in-memory
merge and only instead updates the working copy afterwards (if
requested).

There's still no significant speedup in the hg repo. It seems that the
cost of updating the working copy is not large enough to make much
difference, and the cost of reading and updating obsmarkers is
relatively high in that repo. A significant part of the time (~35%) is
spent in `repoview.computehidden()` because it is recalculated for
every commit (seems like potential for improvement).

I made similar changes in the mozilla-unified repo (which has a
significantly larger working copy), by adding 10 commits each changing
one line of a file. There, evolving 9 of those commits took 34s before
this patch and 20s after. I measured similar speedups in an internal
repo (9.0s -> 5.2s).

#!/usr/bin/env python
import os
import sys
import subprocess

if len(sys.argv) < 2:
    execname = os.path.basename(sys.argv[0])
    sys.stderr.write("usage: %s CLIENT_ID\n" % execname)

client_id = sys.argv[1]

subprocess.check_call(['hg', 'branch', "--force", "hammer-branch-%s" % client_id])

while True:
    subprocess.check_call([
        'hg', 'commit',
        "--config", "ui.allowemptycommit=yes",
        "--message", "hammer-%s" % client_id,
    ])
    nodeid = subprocess.check_output([
        'hg', 'log', '--rev', '.', '--template', '{node}'
    ])
    subprocess.check_call([
        'hg', 'debugobsolete', ''.join(reversed(nodeid)), nodeid
    ])
    subprocess.check_call(['hg', 'pull'])
    subprocess.check_call(['hg', 'push', '--force'])