[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

RE: 1.5.5 next Wednesday (12/17)

From: Julian Foad <julianfoad_at_btopenworld.com>
Date: Thu, 11 Dec 2008 12:49:00 +0000

Bert, that looks like useful info. Please could you put it in a new
email thread or in the issue tracker. I think this is (part of) issue
#3119 <http://subversion.tigris.org/issues/show_bug.cgi?id=3119>, also
referenced recently in email thread
<http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=982066> "Maven scm tag does not work with Subversion 1.5.x".

- Julian

On Thu, 2008-12-11 at 01:33 +0100, Bert Huijben wrote:
> > -----Original Message-----
> > From: Mark Phippard [mailto:markphip_at_gmail.com]
> > Sent: Thursday, December 11, 2008 12:13 AM
> > To: kmradke_at_rockwellcollins.com
> > Cc: dev_at_subversion.tigris.org
> > Subject: Re: 1.5.5 next Wednesday (12/17)
> >
> > On Wed, Dec 10, 2008 at 6:08 PM, <kmradke_at_rockwellcollins.com> wrote:
> > >
> > > Anyone know if the 1.5 regression that causes a:
> > >
> > > svn cp WC URL
> > >
> > > operation to fail with a "file already exists" error
> > > when performed over http when WC has modified files has
> > > been fixed? I'm not seeing any closure to this thread:
> > >
> > > http://svn.haxx.se/users/archive-2008-10/0829.shtml
> > >
> > > I didn't see an issue being filed or anything in 1.5.x STATUS,
> > > but I have run into this particular problem recently and it
> > > worked in 1.4 (and possibly as late as 1.5.1)
> >
> > AFAIK, even 1.6 is in danger of shipping with that bug still in place.
> > I have not seen anyone attempt to fix it.
>
> For other developers looking at this issue:
>
> r31692 (which was merged to 1.5.x in r31799) is the only change between
> 1.5.0 and 1.5.1 in neon that can explain the regression starting in 1.5.1.
> It touches precisely the code Karl quotes in his mail.
>
> Modified: trunk/subversion/libsvn_ra_neon/commit.c
> URL:
> http://svn.collab.net/viewvc/svn/trunk/subversion/libsvn_ra_neon/commit.c?pa
> threv=31692&r1=31691&r2=31692
> ============================================================================
> ==
> --- trunk/subversion/libsvn_ra_neon/commit.c Tue Jun 10 12:59:16 2008
> (r31691)
> +++ trunk/subversion/libsvn_ra_neon/commit.c Tue Jun 10 13:20:28 2008
> (r31692)
> @@ -1036,8 +1036,8 @@ static svn_error_t * commit_add_file(con
> svn_ra_neon__resource_t *res;
> svn_error_t *err = svn_ra_neon__get_starting_props(&res,
> file->cc->ras,
> - file->rsrc->url,
> NULL,
> - workpool);
> +
> file->rsrc->wr_url,
> + NULL, workpool);
> if (!err)
> {
> /* If the PROPFIND succeeds the file already exists */
>
> =====================================================================
> [[
> Fix issue #2939 over ra_neon: bogus existence check.
>
> See reproduction recipe in the issue to verify this fix -- it uses
> svnmucc. Scenario can't be tested using the normal 'svn' client,
> since it involves doing a directory replacement within a single
> transaction. svnmucc and gvn can create the scenario, however,
> resulting is a revision which (until now) couldn't be svnsync'd over
> http!
>
> * subversion/libsvn_ra_neon/commit.c (commit_add_file):
> do PROPFIND existence check on the *working* url (i.e. the path
> within the transaction), not the public url (which the path in HEAD).
> ]]
>
> [[
> Merge r31692, 31695 from trunk:
>
> * r31692, r31695
> Fix issue #2939 over ra_neon/serf: bogus existence check.
> Votes:
> +1: epg, lgo, Arfrever
> ]]
>
> I would guess that reverting this change will bring us back to the old
> behavior, but it will break issue #2939.
>
>
> Karl's diagnosis continues in mod_dav_svn, but the comment above this block
> of code says:
>
> /* If the parent directory existed before this commit then there may be a
> file with this URL already. We need to ensure such a file does not
> exist, which we do by attempting a PROPFIND. Of course, a
> PROPFIND *should* succeed if this "add" is actually the second
> half of a "replace".
>
> [....]
>
> Looking at this neon change, it looks like it starts checking whether the
> file exists in the transaction and stops checking whether it exists before
> the commit. (I'm trying to understand what the consequences are, but the old
> behavior matches what I would expect reading the comments around this code).
>
> The serf change for the same issue is completely different than the neon
> version. (I don't think r31695 needs further investigation)
>
> Bert

------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=982800
Received on 2008-12-11 13:49:32 CET

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.