cleanup of revlog.group when repository is local revlog.group cached every chunk from the revlog, the behaviour was needed to minimize the roundtrip with old-http. The patch export the information that the repository is local or not from the repository object down to the revlog. Then it uses the workaround for old-http only if the repository is non-local. The memory used server side when pulling goes down to less than 30Mo maximum whereas without the patch more than 160Mo was used when cloning the linux kernel repository. The time used by cloning is roughly the same (although some caching could be implemented if needed): before 110.25user 20.90system 2:52.00elapsed 76%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+708707minor)pagefaults 0swaps after 112.85user 22.98system 2:50.66elapsed 79%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+862862minor)pagefaults 0swaps

     1

# httprangereader.py - just what it says

     2

#

     3

# Copyright 2005 Matt Mackall <mpm@selenic.com>

     4

#

     5

# This software may be used and distributed according to the terms

     6

# of the GNU General Public License, incorporated herein by reference.

     7

     8

import byterange, urllib2

     9

    10

class httprangereader(object):

    11

    def __init__(self, url):

    12

        self.url = url

    13

        self.pos = 0

    14

    def seek(self, pos):

    15

        self.pos = pos

    16

    def read(self, bytes=None):

    17

        opener = urllib2.build_opener(byterange.HTTPRangeHandler())

    18

        urllib2.install_opener(opener)

    19

        req = urllib2.Request(self.url)

    20

        end = ''

    21

        if bytes: end = self.pos + bytes

    22

        req.add_header('Range', 'bytes=%d-%s' % (self.pos, end))

    23

        f = urllib2.urlopen(req)

    24

        return f.read()