You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by mike south <ms...@nandomedia.com> on 2004/11/11 18:21:08 UTC

loss of committed changes

Hi,

we have seen a few incidents where a committed change gets lost, and in 
at least one case lost repeatedly.  In searching the archives for this I 
noticed a post that seemed to be describing the same thing I have seen:

http://svn.haxx.se/dev/archive-2002-04/0008.shtml

relevant text:

 >> You forgot to actually declare dav_svn_get_repo_name.
 >> dav_svn_repo_name is static, so that doesn't belong in the header 
(it >>isn't
 >> in there, but is in the commit msg).

 >Not sure why I have lost that change several times. Fixed though.

 >Ben

If, by saying that, Ben is saying that he committed a change and it 
disappeared, we're seeing that--a lot.  We noticed an isolated case here 
and there in the past, but yesterday when  bunch of developers were 
working, often in the same file, we would see changes disappear quite 
frequently.

Here is the sequence:

     developer A commits, changing file Foo.pl

     developer B updates, getting developer A's changes to Foo.pl,
     running Foo.pl and seeing the results of those changes

     developer B commits, changing file Foo.pl, but not near Developer
     A's changes

     2 lines that Developer A committed are no longer there now, even
     though Developer B did not remove them.

We also had a line get doubled (i.e. the same line of code appears twice 
in the file), and I've seen the change-reverse behavior above with an 
add/subtract pair of lines as well.

In another incident yesterday, Developer A committed, Developer B svn 
upped and saw Developer A's changes come in, Developer C svn upped, but 
didn't see Developer A's changes come in.

Has anyone seen anything like this?  (I would love to provide you with a 
shell script to reproduce, but at this point we don't have it pinned 
down.  It seemed to be a lot more prevalent yesterday with multiple 
people working at once on the same files.)

mike


-- 
-----------------------------------------
       Name : Mike South
      Title : Software Developer
    Company : Nando Media
Work Phone : (919) 861-1259
   Work Fax : (919) 861-1300
Work Email : msouth@nandomedia.com
        AIM : msouthnando
   Web Site : http://www.nandomedia.com/
-----------------------------------------


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: loss of committed changes

Posted by mike south <ms...@nandomedia.com>.
C. Michael Pilato wrote:
> mike south <ms...@nandomedia.com> writes:
> 
> 
>>Has anyone seen anything like this?  (I would love to provide you with
>>a shell script to reproduce, but at this point we don't have it pinned
>>down.  It seemed to be a lot more prevalent yesterday with multiple
>>people working at once on the same files.)
> 
> 
> Could you at least provide the version of Subversion, the backend in
> use (BDB or FSFS), and other such details?
> 
using BDB backend over SSH, redhat boxes.  Below is the output from svn 
--version.  The client and server were built from the same source tree.

/nm/sw/subversion/bin/svn --version
svn, version 1.0.2 (r9423)
    compiled Apr 27 2004, 16:24:38

Copyright (C) 2000-2004 CollabNet.
Subversion is open source software, see http://subversion.tigris.org/
This product includes software developed by CollabNet 
(http://www.Collab.Net/).

The following repository access (RA) modules are available:

* ra_dav : Module for accessing a repository via WebDAV (DeltaV) protocol.
   - handles 'http' schema
   - handles 'https' schema
* ra_local : Module for accessing a repository on local disk.
   - handles 'file' schema
* ra_svn : Module for accessing a repository using the svn network protocol.
   - handles 'svn' schema




-- 
-----------------------------------------
       Name : Mike South
      Title : Software Developer
    Company : Nando Media
Work Phone : (919) 861-1259
   Work Fax : (919) 861-1300
Work Email : msouth@nandomedia.com
        AIM : msouthnando
   Web Site : http://www.nandomedia.com/
-----------------------------------------


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: loss of committed changes

Posted by "C. Michael Pilato" <cm...@collab.net>.
mike south <ms...@nandomedia.com> writes:

> Has anyone seen anything like this?  (I would love to provide you with
> a shell script to reproduce, but at this point we don't have it pinned
> down.  It seemed to be a lot more prevalent yesterday with multiple
> people working at once on the same files.)

Could you at least provide the version of Subversion, the backend in
use (BDB or FSFS), and other such details?

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: loss of committed changes

Posted by kf...@collab.net.
mike south <ms...@nandomedia.com> writes:

> Hi,
> 
> we have seen a few incidents where a committed change gets lost, and
> in at least one case lost repeatedly.  In searching the archives for
> this I noticed a post that seemed to be describing the same thing I
> have seen:
> 
> http://svn.haxx.se/dev/archive-2002-04/0008.shtml

I'm not sure that post is talking about the same thing.  He just says
he's "lost" the change several times, doesn't give any more details.
It doesn't seem like the main point of his mail.

> Here is the sequence:
> 
>      developer A commits, changing file Foo.pl
> 
>      developer B updates, getting developer A's changes to Foo.pl,
>      running Foo.pl and seeing the results of those changes
> 
>      developer B commits, changing file Foo.pl, but not near Developer
>      A's changes
> 
>      2 lines that Developer A committed are no longer there now, even
>      though Developer B did not remove them.
> 
> We also had a line get doubled (i.e. the same line of code appears
> twice in the file), and I've seen the change-reverse behavior above
> with an add/subtract pair of lines as well.
> 
> In another incident yesterday, Developer A committed, Developer B svn
> upped and saw Developer A's changes come in, Developer C svn upped,
> but didn't see Developer A's changes come in.
> 
> Has anyone seen anything like this?  (I would love to provide you with
> a shell script to reproduce, but at this point we don't have it pinned
> down.  It seemed to be a lot more prevalent yesterday with multiple
> people working at once on the same files.)

Wow.  As you can guess, we are salivating for a reproduction recipe
over here.  I've never seen Subversion do anything like this, in four
years of daily use.

You don't say whether Developer B edits Foo.pl before or after
updating to get A's changes.  That might make a difference.

I don't think we can do much without a reproduction recipe.  But, can
you maybe construct one based on the successive revisions of Foo.pl in
your repository and developer's memories of which lines
disappeared/changed when they weren't supposed to?  I would think that
all the information you need is there...

-Karl

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: loss of committed changes

Posted by Gary Feldman <g1...@marsdome.com>.
Ben Collins-Sussman wrote:

> I guess I'm just saying the the probability of a bug of this magnitude 
> existing is extremely tiny, compared to the probability of PEBKAC.  
> Don't take it personally.  :-)

I have to admit that this was my reaction, too.  Let me just add one 
suggestion that I don't think anyone has mentioned yet:

Get a verbose log of the file in question since before the first 
relevant commit.  Whenever I've seen problems like this in the past, 
I've often been able to diagnose it via the log.  One of the more common 
causes is that one person either did a commit that failed, and didn't 
notice, or did a commit of a bunch of files and mistakenly thought that 
the file in question was part of that commit.  Either of these are 
likely if the first expected commit is missing from the log.

Gary


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: loss of committed changes

Posted by Ben Collins-Sussman <su...@collab.net>.
On Nov 11, 2004, at 12:21 PM, mike south wrote:
>  but yesterday when  bunch of developers were working, often in the 
> same file, we would see changes disappear quite frequently.
>

I have to admit, I'm skeptical.  If such a bug existed, it would be so 
incredibly fundamental to svn's core abilities, it should have been 
noticed a hundred times over the course of 4 years of development and 
use.

I'm not pooh-poohing your problem;  I really do believe that you guys 
are seeing changes disappear, and are frustrated.  But historically, 
any time folks have ever complained about something like this, it's 
almost always because of some human error or silly mistake.  For 
example, we seen cases where users get conflicts, get confused about 
how to deal with conflict markers, and toss away other users' changes 
without realizing it.  Or they accidentally copy old versions of files 
on top of working files, again destroying newer changes.  A lot of odd 
things happen.

I guess I'm just saying the the probability of a bug of this magnitude 
existing is extremely tiny, compared to the probability of PEBKAC.  
Don't take it personally.  :-)

The best thing to do is keep a watchful eye... the *instant* something 
suspcicious happens, have your user save the output of shell 'history' 
somewhere... it's the beginning of a transcript that can help us either 
detect PEBKAC or a reproduction recipe.


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: loss of committed changes

Posted by Ni...@sybari.com.
What svn client are you using?  Is there a newer version?
It's possible that your client's merge is the culprit.

Regards,
Nick G





mike south <ms...@nandomedia.com>
11/11/2004 01:21 PM
 
        To:     users@subversion.tigris.org
        cc: 
        Subject:        loss of committed changes


Hi,

we have seen a few incidents where a committed change gets lost, and in 
at least one case lost repeatedly.  In searching the archives for this I 
noticed a post that seemed to be describing the same thing I have seen:

http://svn.haxx.se/dev/archive-2002-04/0008.shtml

relevant text:

 >> You forgot to actually declare dav_svn_get_repo_name.
 >> dav_svn_repo_name is static, so that doesn't belong in the header 
(it >>isn't
 >> in there, but is in the commit msg).

 >Not sure why I have lost that change several times. Fixed though.

 >Ben

If, by saying that, Ben is saying that he committed a change and it 
disappeared, we're seeing that--a lot.  We noticed an isolated case here 
and there in the past, but yesterday when  bunch of developers were 
working, often in the same file, we would see changes disappear quite 
frequently.

Here is the sequence:

     developer A commits, changing file Foo.pl

     developer B updates, getting developer A's changes to Foo.pl,
     running Foo.pl and seeing the results of those changes

     developer B commits, changing file Foo.pl, but not near Developer
     A's changes

     2 lines that Developer A committed are no longer there now, even
     though Developer B did not remove them.

We also had a line get doubled (i.e. the same line of code appears twice 
in the file), and I've seen the change-reverse behavior above with an 
add/subtract pair of lines as well.

In another incident yesterday, Developer A committed, Developer B svn 
upped and saw Developer A's changes come in, Developer C svn upped, but 
didn't see Developer A's changes come in.

Has anyone seen anything like this?  (I would love to provide you with a 
shell script to reproduce, but at this point we don't have it pinned 
down.  It seemed to be a lot more prevalent yesterday with multiple 
people working at once on the same files.)

mike


-- 
-----------------------------------------
       Name : Mike South
      Title : Software Developer
    Company : Nando Media
Work Phone : (919) 861-1259
   Work Fax : (919) 861-1300
Work Email : msouth@nandomedia.com
        AIM : msouthnando
   Web Site : http://www.nandomedia.com/
-----------------------------------------


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org




---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org