Problem regarding hardlink when clone from linux and commit from windows

[Jerome Allasia]

Hi,
Thank you for the fast response ;-)
"mapping the linux drive on V:" => we are using a NAS system : mount with 
NFS on linux box, and using CIFS for windows => 
http://en.wikipedia.org/wiki/Network-attached_storage 

> As a workaround you can use "hg clone -U" or simply "cp -ar" instead.
> hg clone --pull

we are going to use this kind of solution. It's sad not to be able to use 
Hardlinks ... it would have been really nice for us ... anyway i will have 
a look also on my side "inside" to see if i can find a solution for that. 

> What does that comment mean? Are you or are you not using 1.6.4? Can you 

> post the output of "V:\HardlinkTEST\cloneRef>hg version"?

V:\HardlinkTEST\cloneRef>hg version
Mercurial Distributed SCM (version 1.6.1023)

Copyright (C) 2005-2010 Matt Mackall <mpm at selenic.com> and others
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR 
PURPOSE.

> For issue761, there should be workaround released with 1.6.3. in effect.
> But that applies to Windows shares only.

Do you think that we can extends/activate that fix on demand (having an 
option to the commit to force it to consider all files as hardlinks) ?




Adrian Buehlmann <adrian at cadifra.com> 
To
Jerome Allasia <jerome.allasia at amadeus.com>
cc
mercurial at selenic.com
Subject
Re: Problem regarding hardlink when clone from linux and commit from 
windows





Adrian Buehlmann <adrian at cadifra.com> 
21/10/2010 17:18


On 21.10.2010 15:54, Jerome Allasia wrote:
> 
> Hi,
> 
> The problem is the following (i have read
> 
http://mercurial.selenic.com/wiki/HardlinkedClones#Hardlinked_clones_on_Windows_shares
,
> but it seams that the problem still exist ...)
> 
>     * I have a repository call "ref", first i clone it to cloneRef (cmd
>       run on the linux machine, on a linux terminal)

This will create a so called "hardlinked clone" as described on that
wiki page.

> jallasia at ncegcolnx123:~/HardlinkTEST) hg clone ref cloneRef 
> 
> /jallasia at ncegcolnx123:~/HardlinkTEST) ll 
>                                                    1:03PM/
> /total 16/
> /drwxr-xr-x 3 jallasia dev_air 1024 Oct 21 11:17 cloneRef/
> /drwxr-xr-x 3 jallasia dev_air 1024 Oct 20 12:24 ref/
> 
>     * I modify a file call "test.txt" inside cloneRef
>     * On windows i perform a commit : (run on windows, mapping the linux
>       drive on V: using mercurial v1.6.4)

Re: "mapping the linux drive on V:"

How is that done? Are you running a Samba server, and then sharing the
files like a Windows share, accessing it from Windows as clients?

Or?

> V:\HardlinkTEST\cloneRef>hg commit
> 
>     * the result it that both repository are modified !!!

Which is wrong and a strong sign that hardlinks are not broken up by the
commit. Which may be caused by mercurial getting told a wrong hardlink
count when it asks for the number of hardlinks for a file.

> jallasia at ncegcolnx123:~/HardlinkTEST) ll cloneRef/.hg/store/data
> ref/.hg/store/data                                       1:09PM
> cloneRef/.hg/store/data:
> total 8
> -rw-r--r-- 2 jallasia dev_air 223 Oct 21 11:17 test.txt.i
> 
> ref/.hg/store/data:
> total 8
> -rw-r--r-- 2 jallasia dev_air 223 Oct 21 11:17 test.txt.i
> 
> (I have install the version 1.6.4 on windows to perform that ...)
> 
> Do you have a solution for me ????

Not until we know more. See my question above. You might get faster,
interactive response on IRC channel #mercurial (my nick is abuehl if I'm
on the channel).

BTW 'hg clone --pull' will create clones that have no hardlinks. But
there is no reliable way to enforce that.

> (THIS PROBLEM IS REALLY A SHOW STOPER ! AN WE HAVE POTENTIALLY HUNDREDS
> OF DEVELOPPER IN OUR COMPANIE THAT ARE OR WILL USE IT OR MIGRATE TO
> ANOTHER SOLUTION)

No need to use BIG CAPS. There are quite a number of volunteer people on
the mercurial project, helping out unpaid in their free time. Screaming
at us isn't going to be helpful in any way.

The problem reports we had so far regarding problems with hardlinks:

* http://mercurial.selenic.com/bts/issue761
* http://mercurial.selenic.com/bts/issue1866

None of which were caused by bugs in mercurial itself (either windows
failing to report the hardlink count or a Linux driver having similar
problems).

For issue761, there should be workaround released with 1.6.3. in effect.
But that applies to Windows shares only.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mercurial-scm.org/pipermail/mercurial/attachments/20101021/e211f0b1/attachment.html>