You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Eelco Dolstra <ee...@cs.uu.nl> on 2003/03/04 16:23:49 UTC

mod_dav_svn backward compatibility broken?

Hi,

I just tried to install the latest Subversion from the trunk (revision
5197) on our repository server, but it appears that older clients (0.18)
are unable to do updates from the newer repository server:

$ svn co http://losser.st-lab.cs.uu.nl:12080/repos/test/
A  test/D
...
Checked out revision 25.

$ cd test/

$ svn up
svn: RA layer request failed
svn: REPORT request failed on /repos/test
svn: The REPORT request returned invalid XML in the response: Unknown
XML element `repository-uuid (in
http://subversion.tigris.org/xmlns/dav/)'. (/repos/test)

$ svn --version
svn, version 0.18.0 (r4973)
   compiled Feb 25 2003, 12:42:35

So checkouts still work, but updates are broken.  Is this the intended
behaviour?

FWIW, the server is running FreeBSD 4.7, the client is SuSE 8.1.

Eelco.

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

Re: mod_dav_svn backward compatibility broken?

Posted by Karl Fogel <kf...@newton.ch.collab.net>.
Branko Čibej <br...@xbc.nu> writes:
> Why can't we jiust marshal the uuid as a property that just happens to
> be present on every versionable resource? That not only makes sense
> semantically (since we should be storing the uuid in the entries file),
> it also means that any DAV client will see it, but can ignore it.

I think that's what we're doing right now... (?)
 

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

Re: mod_dav_svn backward compatibility broken?

Posted by Branko Čibej <br...@xbc.nu>.
mark benedetto king wrote:

>4.) Use existing properties infrastructure.
>
>	The only thing I don't like about this is that the other WC
>	props are transmitted one way, why should this one be different?
>
It shouldn't -- the *others* should become like this one. :-)

>I'm not married to any one of these proposals; I think they all have
>merit.  If I had to choose, I think (4) would get my vote.
>
Mine, too.

-- 
Brane Čibej   <br...@xbc.nu>   http://www.xbc.nu/brane/


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

Re: mod_dav_svn backward compatibility broken?

Posted by Greg Stein <gs...@lyra.org>.
On Tue, Mar 04, 2003 at 06:49:10PM -0500, mark benedetto king wrote:
>...
> On Tue, Mar 04, 2003 at 02:49:49PM -0800, Greg Stein wrote:
> > 
> >   1) stop returning NE_XML_DECLINE. return NE_XML_VALID and jettison the
> >      unknown element in our start_handler. (this shifts the "ignore unknown
> >      elements" from the validator to the element handler)
> > 
> >   3) beat Joe into adding NE_XML_IGNORE for elements which are "valid" but
> >      can simply be ignored [if no other handler wants them]. we would then
> >      need a mandated upgrade of Neon.
>...
> a.) immediately stop sending UUIDs on update.  I *think* the liveprop
>     entries are okay.  Are they?

Dunno.

> b.) in the same atomic change, start sending an "X-SVN-Repository-UUID: "
>     header, and modify ra_dav to handle this.

As long as this is temporary and will be removed.

> c.) do (1)
> d.) do (3)

and revert (c).

> e.) wait until most clients are updated.  I think, at a minimum, this means
>     binary packages available for most OSen + one more month.  There's really
>     no forcing function on this.
> f.) revert (a)

revert (a) *and* (b).

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/

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

Re: mod_dav_svn backward compatibility broken?

Posted by Ben Collins-Sussman <su...@collab.net>.
mark benedetto king <mb...@boredom.org> writes:

> On Tue, Mar 04, 2003 at 02:49:49PM -0800, Greg Stein wrote:
> > 
> >   1) stop returning NE_XML_DECLINE. return NE_XML_VALID and jettison the
> >      unknown element in our start_handler. (this shifts the "ignore unknown
> >      elements" from the validator to the element handler)
> > 
> >   3) beat Joe into adding NE_XML_IGNORE for elements which are "valid" but
> >      can simply be ignored [if no other handler wants them]. we would then
> >      need a mandated upgrade of Neon.
> > 
> > My favorite is option (1) for short term (the 0.19 release), and option (3)
> > for the long term plan.
> > 
> > In all cases, this also implies that the server should stop sending the UUID
> > for now, until we get "enough" clients with the flexible handling. If we
> > really need the server to send the UUID, then we can *temporarily* put the
> > thing into the server headers; that works fine for ra_dav, but is not a good
> > long-term WebDAV-y solution.
> > 
> 
> Why don't we do this:
> 
> a.) immediately stop sending UUIDs on update.  I *think* the liveprop
>     entries are okay.  Are they?
> b.) in the same atomic change, start sending an "X-SVN-Repository-UUID: "
>     header, and modify ra_dav to handle this.
> c.) do (1)
> d.) do (3)
> e.) wait until most clients are updated.  I think, at a minimum, this means
>     binary packages available for most OSen + one more month.  There's really
>     no forcing function on this.
> f.) revert (a)
> 
> Thoughts?

+1.  Amen, Brother.

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

Re: mod_dav_svn backward compatibility broken?

Posted by mark benedetto king <mb...@boredom.org>.
On Tue, Mar 04, 2003 at 02:49:49PM -0800, Greg Stein wrote:
> 
>   1) stop returning NE_XML_DECLINE. return NE_XML_VALID and jettison the
>      unknown element in our start_handler. (this shifts the "ignore unknown
>      elements" from the validator to the element handler)
> 
>   3) beat Joe into adding NE_XML_IGNORE for elements which are "valid" but
>      can simply be ignored [if no other handler wants them]. we would then
>      need a mandated upgrade of Neon.
> 
> My favorite is option (1) for short term (the 0.19 release), and option (3)
> for the long term plan.
> 
> In all cases, this also implies that the server should stop sending the UUID
> for now, until we get "enough" clients with the flexible handling. If we
> really need the server to send the UUID, then we can *temporarily* put the
> thing into the server headers; that works fine for ra_dav, but is not a good
> long-term WebDAV-y solution.
> 

Why don't we do this:

a.) immediately stop sending UUIDs on update.  I *think* the liveprop
    entries are okay.  Are they?
b.) in the same atomic change, start sending an "X-SVN-Repository-UUID: "
    header, and modify ra_dav to handle this.
c.) do (1)
d.) do (3)
e.) wait until most clients are updated.  I think, at a minimum, this means
    binary packages available for most OSen + one more month.  There's really
    no forcing function on this.
f.) revert (a)

Thoughts?

--ben

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

Re: mod_dav_svn backward compatibility broken?

Posted by Greg Stein <gs...@lyra.org>.
On Tue, Mar 04, 2003 at 03:16:01PM -0500, mark benedetto king wrote:
>...
> 1.) A custom request header
>...
> 2.) A capabilities header
>...
> 3.) A custom *response* header

These are all way, way bogus. Non-starters.

>...
> 4.) Use existing properties infrastructure.

I'm unclear on what this means, but am skipping anyways.

>...
> 5.) Do nothing.

Doesn't work in the long run.

>...
> 6.) Do less than nothing (stop validating S:prop so thoroughly) (be
>     liberal in what we accept).

The only valid answer.

We have to be able to deal with new elements suddenly appearing at any place
in the XML response. Otherwise, we'll never get any kind of interoperability
moving forward. If the server starts stuffing new items in, then the client
should just ignore them. If the client expects something, but the server
doesn't provide it, then the client should gracefully degrade (or signal
that it requires the new data from the server, and cleanly abort the
operation).


There are several things that we can do:

  1) stop returning NE_XML_DECLINE. return NE_XML_VALID and jettison the
     unknown element in our start_handler. (this shifts the "ignore unknown
     elements" from the validator to the element handler)

  2) stop using Neon's XML parsing. we have Expat, so this might not be too
     hard, but is certainly more work than (1).

  3) beat Joe into adding NE_XML_IGNORE for elements which are "valid" but
     can simply be ignored [if no other handler wants them]. we would then
     need a mandated upgrade of Neon.

My favorite is option (1) for short term (the 0.19 release), and option (3)
for the long term plan.

In all cases, this also implies that the server should stop sending the UUID
for now, until we get "enough" clients with the flexible handling. If we
really need the server to send the UUID, then we can *temporarily* put the
thing into the server headers; that works fine for ra_dav, but is not a good
long-term WebDAV-y solution.

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/

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

Re: mod_dav_svn backward compatibility broken?

Posted by Joe Orton <jo...@manyfish.co.uk>.
On Tue, Mar 04, 2003 at 03:16:01PM -0500, mark benedetto king wrote:
> On Tue, Mar 04, 2003 at 01:53:15PM -0800, Ben Collins-Sussman wrote:
> > Branko Čibej <br...@xbc.nu> writes:
> > 
> > > Why can't we jiust marshal the uuid as a property that just happens to
> > > be present on every versionable resource? That not only makes sense
> > > semantically (since we should be storing the uuid in the entries file),
> > > it also means that any DAV client will see it, but can ignore it.
> > 
> > Um, yes, that's exactly what we're doing.  If you use WebDAV to ask a
> > resource for all its properties, you'll see the UUID, DAV props, and
> > user-controlled props.
> 
> Right.  I think what Brane was getting at is that it should have been
> shipped as one of the user-controlled props (which obviously don't suffer
> from backwards-compatibility issues).  I don't see anything wrong with that,
> but I did include it where I did because it is processed in much the same
> way.
>
> Proposed so far to resolve this have been:

Another hack might be to include the UUID as an attribute on a property
which is accepted by 0.18 clients?

I apologise for the ne_xml.h interface, one of its many sins is that it
encourages the application to validate responses, which is pretty
irresponsible.

joe

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

Re: mod_dav_svn backward compatibility broken?

Posted by Karl Fogel <kf...@newton.ch.collab.net>.
mark benedetto king <mb...@boredom.org> writes:
> 3.) A custom *response* header
> 
> 	X-Repository-UUID: 32e0795c-b79d-4482-a8e7-4bf008f4a25f
> 
> 	This is probably how I would have done it in the first place, had
> 	it occurred to me, but it does preclude sending different UUIDs
> 	for different files.

Does it really preclude that?  What if the header UUID is treated as
the default, but an individual resource could override it, someday in
the future.

> 5.) Do nothing.
> 
> 	It's alpha, we can break the protocol.  However, if we upgrade
> 	the svn.collab.net server and prevent everyone from upgrading
> 	with svn up, that would probably be bad, if only in terms of
> 	the increased dev@ list volume.

We'd wait to upgrade, if it were going to be incompatible.

> 6.) Do less than nothing (stop validating S:prop so thoroughly) (be
>     liberal in what we accept).
> 
> 	Has all the problems of (5), but solves some things Once and
> 	For All(tm).  

This isn't really an option for the present problem, because it
doesn't help old clients (0.18.0 and 0.18.1).  We can't go back and
change *their* code :-).

> I'm not married to any one of these proposals; I think they all have
> merit.  If I had to choose, I think (4) would get my vote.

Me too.

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

Re: mod_dav_svn backward compatibility broken?

Posted by mark benedetto king <mb...@boredom.org>.
On Tue, Mar 04, 2003 at 01:53:15PM -0800, Ben Collins-Sussman wrote:
> Branko Čibej <br...@xbc.nu> writes:
> 
> > Why can't we jiust marshal the uuid as a property that just happens to
> > be present on every versionable resource? That not only makes sense
> > semantically (since we should be storing the uuid in the entries file),
> > it also means that any DAV client will see it, but can ignore it.
> 
> Um, yes, that's exactly what we're doing.  If you use WebDAV to ask a
> resource for all its properties, you'll see the UUID, DAV props, and
> user-controlled props.

Right.  I think what Brane was getting at is that it should have been
shipped as one of the user-controlled props (which obviously don't suffer
from backwards-compatibility issues).  I don't see anything wrong with that,
but I did include it where I did because it is processed in much the same
way.

Proposed so far to resolve this have been:

1.) A custom request header

	X-I-Know-About-UUIDs: true

	This would be easy to add, I think, but it might be worth
	doing (2) instead.

2.) A capabilities header

	X-I-Know-About: UUIDs, svndiff2 

	This would be only marginally more difficult than (1), but does
	it open a slippery slope?  Personally, I like it more than something
	like "X-RA-DAV-Version: 2", so that clients need not be fully
	compliant.

3.) A custom *response* header

	X-Repository-UUID: 32e0795c-b79d-4482-a8e7-4bf008f4a25f

	This is probably how I would have done it in the first place, had
	it occurred to me, but it does preclude sending different UUIDs
	for different files.

4.) Use existing properties infrastructure.

	The only thing I don't like about this is that the other WC
	props are transmitted one way, why should this one be different?

I suppose there are also:

5.) Do nothing.

	It's alpha, we can break the protocol.  However, if we upgrade
	the svn.collab.net server and prevent everyone from upgrading
	with svn up, that would probably be bad, if only in terms of
	the increased dev@ list volume.

6.) Do less than nothing (stop validating S:prop so thoroughly) (be
    liberal in what we accept).

	Has all the problems of (5), but solves some things Once and
	For All(tm).  


I'm not married to any one of these proposals; I think they all have
merit.  If I had to choose, I think (4) would get my vote.

--ben


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

Re: mod_dav_svn backward compatibility broken?

Posted by Branko Čibej <br...@xbc.nu>.
mark benedetto king wrote:

>On Tue, Mar 04, 2003 at 09:00:28PM +0100, Branko ?ibej wrote:
>  
>
>>Aha. So our custom report language (or its parser) is supid enough to
>>not ignore "unknown" elements. Sigh, sigh, sigh.
>>
>>    
>>
>
>It's not stupid; it's *designed* not to ignore them.
>
O.K., it's not stupid /enough/. IMHO, a report language like this should
be at least marginally extensible, or tolerant of some kinds of unknown
records, so that older clients can interact with newer servers without
*too* many problems.


-- 
Brane Čibej   <br...@xbc.nu>   http://www.xbc.nu/brane/


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

Re: mod_dav_svn backward compatibility broken?

Posted by mark benedetto king <mb...@boredom.org>.
On Tue, Mar 04, 2003 at 09:00:28PM +0100, Branko ?ibej wrote:
> Aha. So our custom report language (or its parser) is supid enough to
> not ignore "unknown" elements. Sigh, sigh, sigh.
> 

It's not stupid; it's *designed* not to ignore them.

I address this issue in my mail on this thread 10 seconds ago.

--ben


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

Re: mod_dav_svn backward compatibility broken?

Posted by Branko Čibej <br...@xbc.nu>.
Ben Collins-Sussman wrote:

>Branko Čibej <br...@xbc.nu> writes:
>
>  
>
>>Why can't we jiust marshal the uuid as a property that just happens to
>>be present on every versionable resource? That not only makes sense
>>semantically (since we should be storing the uuid in the entries file),
>>it also means that any DAV client will see it, but can ignore it.
>>    
>>
>
>Um, yes, that's exactly what we're doing.  If you use WebDAV to ask a
>resource for all its properties, you'll see the UUID, DAV props, and
>user-controlled props.
>
>The problem is that our "custom" update report returns the
>non-user-controlled props within the tree delta -- on each resource.
>Here's a bit of the tree-delta sent back during an update:
>
><S:add-file name="python_3bits.txt"
>result-checksum="ec816060540c0da8ea6c882cf470913c" >
>
><D:checked-in><D:href>/svn/repo4/!svn/ver/1/3bits/python_3bits.txt</D:href>
></D:checked-in>
>
><S:prop>
><D:version-name>1</D:version-name>
><D:creationdate>2003-03-04T18:40:09.748608Z</D:creationdate>
><D:creator-displayname>sussman</D:creator-displayname>
><S2:repository-uuid>cb4d11d7-0fb8-0310-b2e3-bd6612ce3422
></S2:repository-uuid>
></S:prop>
>
></S:add-file>
>
>And older ra_dav libraries are choking, because they don't think the
><repostiory-uuid> element is valid element in the custom report
>language.  Grrrr.
>  
>
Aha. So our custom report language (or its parser) is supid enough to
not ignore "unknown" elements. Sigh, sigh, sigh.


-- 
Brane Čibej   <br...@xbc.nu>   http://www.xbc.nu/brane/


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

Re: mod_dav_svn backward compatibility broken?

Posted by Ben Collins-Sussman <su...@collab.net>.
Branko Čibej <br...@xbc.nu> writes:

> Why can't we jiust marshal the uuid as a property that just happens to
> be present on every versionable resource? That not only makes sense
> semantically (since we should be storing the uuid in the entries file),
> it also means that any DAV client will see it, but can ignore it.

Um, yes, that's exactly what we're doing.  If you use WebDAV to ask a
resource for all its properties, you'll see the UUID, DAV props, and
user-controlled props.

The problem is that our "custom" update report returns the
non-user-controlled props within the tree delta -- on each resource.
Here's a bit of the tree-delta sent back during an update:

<S:add-file name="python_3bits.txt"
result-checksum="ec816060540c0da8ea6c882cf470913c" >

<D:checked-in><D:href>/svn/repo4/!svn/ver/1/3bits/python_3bits.txt</D:href>
</D:checked-in>

<S:prop>
<D:version-name>1</D:version-name>
<D:creationdate>2003-03-04T18:40:09.748608Z</D:creationdate>
<D:creator-displayname>sussman</D:creator-displayname>
<S2:repository-uuid>cb4d11d7-0fb8-0310-b2e3-bd6612ce3422
</S2:repository-uuid>
</S:prop>

</S:add-file>

And older ra_dav libraries are choking, because they don't think the
<repostiory-uuid> element is valid element in the custom report
language.  Grrrr.

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

Re: mod_dav_svn backward compatibility broken?

Posted by Branko Čibej <br...@xbc.nu>.
Why can't we jiust marshal the uuid as a property that just happens to
be present on every versionable resource? That not only makes sense
semantically (since we should be storing the uuid in the entries file),
it also means that any DAV client will see it, but can ignore it.

Ben Collins-Sussman wrote:

>Karl Fogel <kf...@newton.ch.collab.net> writes:
>
>  
>
>>Ben Collins-Sussman <su...@collab.net> writes:
>>    
>>
>>>But we can't retroactively change released clients.  mbk, we need to
>>>come up with a solution here.  Perhaps a different, more compatible
>>>way to marshal the UUID from server to client.  We can't release a
>>>0.19 server that chokes earlier clients.  :-(
>>>      
>>>
>>I'm not sure about "can't" (we can always warn people to upgrade
>>clients before servers, this being alpha and all).  But it would sure
>>be nice to avoid it!
>>
>>MBK, one solution might be to add a header to the client requests,
>>something like
>>
>>   X-repository-uuid-accepted
>>
>>...or whatever.  The server would never send UUIDs unless this header
>>were present.  But is this veering dangerously close to a full
>>features exchange, implying that we should come up with a more general
>>mechanism for that?
>>    
>>
>
>Or even easier: have a UUID-aware server add an X-SVN-Repository-UUID:
>header to its responses.  An ignorant client will just ignore the
>header.  A new client will explicitly look for the header.
>
>Note that this means the UUID is global to the server repsonse, not
>attached to individual files.  That's fine for the moment.  We can
>teach 0.19 to not choke on per-file UUID properties, but we won't send
>them just yet.  In the future, they can be used to 'override' the
>global value, much like the way our entries file works.
>
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
>For additional commands, e-mail: dev-help@subversion.tigris.org
>
>  
>


-- 
Brane Čibej   <br...@xbc.nu>   http://www.xbc.nu/brane/


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

Re: mod_dav_svn backward compatibility broken?

Posted by Ben Collins-Sussman <su...@collab.net>.
Karl Fogel <kf...@newton.ch.collab.net> writes:

> Ben Collins-Sussman <su...@collab.net> writes:
> > But we can't retroactively change released clients.  mbk, we need to
> > come up with a solution here.  Perhaps a different, more compatible
> > way to marshal the UUID from server to client.  We can't release a
> > 0.19 server that chokes earlier clients.  :-(
> 
> I'm not sure about "can't" (we can always warn people to upgrade
> clients before servers, this being alpha and all).  But it would sure
> be nice to avoid it!
> 
> MBK, one solution might be to add a header to the client requests,
> something like
> 
>    X-repository-uuid-accepted
> 
> ...or whatever.  The server would never send UUIDs unless this header
> were present.  But is this veering dangerously close to a full
> features exchange, implying that we should come up with a more general
> mechanism for that?

Or even easier: have a UUID-aware server add an X-SVN-Repository-UUID:
header to its responses.  An ignorant client will just ignore the
header.  A new client will explicitly look for the header.

Note that this means the UUID is global to the server repsonse, not
attached to individual files.  That's fine for the moment.  We can
teach 0.19 to not choke on per-file UUID properties, but we won't send
them just yet.  In the future, they can be used to 'override' the
global value, much like the way our entries file works.


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

Re: mod_dav_svn backward compatibility broken?

Posted by Karl Fogel <kf...@newton.ch.collab.net>.
Ben Collins-Sussman <su...@collab.net> writes:
> But we can't retroactively change released clients.  mbk, we need to
> come up with a solution here.  Perhaps a different, more compatible
> way to marshal the UUID from server to client.  We can't release a
> 0.19 server that chokes earlier clients.  :-(

I'm not sure about "can't" (we can always warn people to upgrade
clients before servers, this being alpha and all).  But it would sure
be nice to avoid it!

MBK, one solution might be to add a header to the client requests,
something like

   X-repository-uuid-accepted

...or whatever.  The server would never send UUIDs unless this header
were present.  But is this veering dangerously close to a full
features exchange, implying that we should come up with a more general
mechanism for that?

-K

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

Re: mod_dav_svn backward compatibility broken?

Posted by Ben Collins-Sussman <su...@collab.net>.
Eelco Dolstra <ee...@cs.uu.nl> writes:

> $ svn up
> svn: RA layer request failed
> svn: REPORT request failed on /repos/test
> svn: The REPORT request returned invalid XML in the response: Unknown
> XML element `repository-uuid (in
> http://subversion.tigris.org/xmlns/dav/)'. (/repos/test)

Ouch, this is a big problem.  After release 0.18.0 and 0.18.1, we
rolled the UUID stuff into the trunk.  The server now adds

<S2:repository-uuid>cb4d11d7-0fb8-0310-b2e3-bd6612ce3422</S2:repository-uuid>

elements within the update reports, which aren't understood by
anything but HEAD clients.  libsvn_ra_dav is validating the report
language, and chokes on the new xml element around line 1723 or so:

 case ELEM_prop:
      if (child == ELEM_version_name
          || child == ELEM_creationdate
          || child == ELEM_creator_displayname
          || child == ELEM_md5_checksum
          || child == ELEM_repository_uuid  /** Not present in 0.18 clients! */
          || child == ELEM_remove_prop)
        return NE_XML_VALID;
      else
        return NE_XML_DECLINE;

We're being bitten by our validator, which is being too strict, argh.

But we can't retroactively change released clients.  mbk, we need to
come up with a solution here.  Perhaps a different, more compatible
way to marshal the UUID from server to client.  We can't release a
0.19 server that chokes earlier clients.  :-(


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