You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Hyrum K Wright <hy...@wandisco.com> on 2012/03/19 21:48:03 UTC

ra_serf not providing a valid revision to delete_entry() upon update

[ warning: investigation is still ongoing, but I thought I'd report this here.]

I'm trying to debug the Ev2 shims over ra_dav.  In doing so, I've
discovered an inconsistency between ra_serf and ra_neon (surprise!)

ra_neon provides a valid argument for the replaced_rev parameter of
the Ev1 delete_entry() function.  The relevant code is in
libsvn_ra_neon/replay.c:

169         const char *path = svn_xml_get_attr_value("name", atts);
170         const char *crev = svn_xml_get_attr_value("rev", atts);
171
172         if (! path)
173           return MISSING_ATTR(nspace, elt_name, "name");
174         else if (! crev)
175           return MISSING_ATTR(nspace, elt_name, "rev");
176         else
177           {
178             dir_item_t *di = &TOP_DIR(rb);
179
180             SVN_ERR(rb->editor->delete_entry(path, SVN_STR_TO_REV(crev),
181                                              di->baton, di->pool));
182           }
183       }
184       break;

Apparently, ra_neon expects and finds a "rev" attribute on the
relevant XML element.  However, the same code in ra_serf (update.c)
doesn't:

1556   else if ((state == OPEN_DIR || state == ADD_DIR) &&
1557            strcmp(name.name, "delete-entry") == 0)
1558     {
1559       const char *file_name;
1560       report_info_t *info;
1561       apr_pool_t *tmppool;
1562       const char *full_path;
1563
1564       file_name = svn_xml_get_attr_value("name", attrs);
1565
1566       if (!file_name)
1567         {
1568           return svn_error_create(
1569             SVN_ERR_RA_DAV_MALFORMED_DATA, NULL,
1570             _("Missing name attr in delete-entry element"));
1571         }
1572
1573       info = parser->state->private;
1574
1575       SVN_ERR(open_dir(info->dir));
1576
1577       tmppool = svn_pool_create(info->dir->dir_baton_pool);
1578
1579       full_path = svn_relpath_join(info->dir->name, file_name, tmppool);
1580
1581       SVN_ERR(info->dir->update_editor->delete_entry(full_path,
1582                                                      SVN_INVALID_REVNUM,
1583                                                      info->dir->dir_baton,
1584                                                      tmppool));
1585
1586       svn_pool_destroy(tmppool);
1587     }

As you can see, ra_serf is unconditionally passing SVN_INVALID_REVNUM
to delete_entry(), which is what is causing the shims to fail over
ra_dav (when run with ra_serf).

I tried inserting analogous code to fetch the "rev" attribute in
ra_serf, but it appears that said attribute isn't available.  I'm not
familiar enough with the protocol to know why it would be for ra_neon,
but not ra_serf, and I'm asking here in the hopes that somebody can
enlighten me. :)

-Hyrum

-- 

uberSVN: Apache Subversion Made Easy
http://www.uberSVN.com/

Re: ra_serf not providing a valid revision to delete_entry() upon update

Posted by Hyrum K Wright <hy...@wandisco.com>.
On Mon, Mar 19, 2012 at 5:24 PM, Greg Stein <gs...@gmail.com> wrote:
> On Mon, Mar 19, 2012 at 18:16, Greg Stein <gs...@gmail.com> wrote:
>> On Mon, Mar 19, 2012 at 16:48, Hyrum K Wright <hy...@wandisco.com> wrote:
>>> [ warning: investigation is still ongoing, but I thought I'd report this here.]
>>>
>>> I'm trying to debug the Ev2 shims over ra_dav.  In doing so, I've
>>> discovered an inconsistency between ra_serf and ra_neon (surprise!)
>>
>> ra_neon uses a "replay-report" while ra_serf uses an "update-report".
>> The former provides the rev= attribute.
>>
>> I don't know why Neon uses the replay-report instead of the
>> update-report. I dunno what the differences are.
>
> What operations are you using? And why is it different between the two RA runs?

I see this on history test 1 (among others) when running davautocheck
and compiled with the shims enabled.  It happens run running 'svn up
-r1' when the working copy is at r2, and r2 had a file replacement.

I might have been wrong about which bit of ra_neon was driving the
delete_entry() handler.  I'll double check that bit later this
evening.

-Hyrum

> Note: in neon/fetch.c:1845, I see that SVN_INVALID_REVNUM is passed to
> delete_entry(). That happens when ra_neon runs an update-report.
>
> Cheers,
> -g



-- 

uberSVN: Apache Subversion Made Easy
http://www.uberSVN.com/

Re: ra_serf not providing a valid revision to delete_entry() upon update

Posted by Hyrum K Wright <hy...@wandisco.com>.
On Mon, Mar 19, 2012 at 8:03 PM, Hyrum K Wright
<hy...@wandisco.com> wrote:
> On Mon, Mar 19, 2012 at 5:24 PM, Greg Stein <gs...@gmail.com> wrote:
>> On Mon, Mar 19, 2012 at 18:16, Greg Stein <gs...@gmail.com> wrote:
>>> On Mon, Mar 19, 2012 at 16:48, Hyrum K Wright <hy...@wandisco.com> wrote:
>>>> [ warning: investigation is still ongoing, but I thought I'd report this here.]
>>>>
>>>> I'm trying to debug the Ev2 shims over ra_dav.  In doing so, I've
>>>> discovered an inconsistency between ra_serf and ra_neon (surprise!)
>>>
>>> ra_neon uses a "replay-report" while ra_serf uses an "update-report".
>>> The former provides the rev= attribute.
>>>
>>> I don't know why Neon uses the replay-report instead of the
>>> update-report. I dunno what the differences are.
>>
>> What operations are you using? And why is it different between the two RA runs?
>>
>> Note: in neon/fetch.c:1845, I see that SVN_INVALID_REVNUM is passed to
>> delete_entry(). That happens when ra_neon runs an update-report.
>
> You are correct, this code path uses neon/fetch.c:1845 on the ra_neon
> side, not the snippet I cited earlier.
>
> So the information is getting dropped over *both* ra_dav clients, not
> just ra_serf.  This is still a disconnect from that which is provided
> directly to the delete_entry() callback from the repository in
> reporter.c:916.  This bit of data is getting dropped.

The reason this bit of data was being dropped is that the delta editor
defines this as an optional piece of data. :(  r1302758 attempts to
paper around the issue in the compat shims.

-Hyrum


-- 

uberSVN: Apache Subversion Made Easy
http://www.uberSVN.com/

Re: ra_serf not providing a valid revision to delete_entry() upon update

Posted by Hyrum K Wright <hy...@wandisco.com>.
On Mon, Mar 19, 2012 at 5:24 PM, Greg Stein <gs...@gmail.com> wrote:
> On Mon, Mar 19, 2012 at 18:16, Greg Stein <gs...@gmail.com> wrote:
>> On Mon, Mar 19, 2012 at 16:48, Hyrum K Wright <hy...@wandisco.com> wrote:
>>> [ warning: investigation is still ongoing, but I thought I'd report this here.]
>>>
>>> I'm trying to debug the Ev2 shims over ra_dav.  In doing so, I've
>>> discovered an inconsistency between ra_serf and ra_neon (surprise!)
>>
>> ra_neon uses a "replay-report" while ra_serf uses an "update-report".
>> The former provides the rev= attribute.
>>
>> I don't know why Neon uses the replay-report instead of the
>> update-report. I dunno what the differences are.
>
> What operations are you using? And why is it different between the two RA runs?
>
> Note: in neon/fetch.c:1845, I see that SVN_INVALID_REVNUM is passed to
> delete_entry(). That happens when ra_neon runs an update-report.

You are correct, this code path uses neon/fetch.c:1845 on the ra_neon
side, not the snippet I cited earlier.

So the information is getting dropped over *both* ra_dav clients, not
just ra_serf.  This is still a disconnect from that which is provided
directly to the delete_entry() callback from the repository in
reporter.c:916.  This bit of data is getting dropped.

-Hyrum


-- 

uberSVN: Apache Subversion Made Easy
http://www.uberSVN.com/

Re: ra_serf not providing a valid revision to delete_entry() upon update

Posted by Greg Stein <gs...@gmail.com>.
On Mon, Mar 19, 2012 at 18:16, Greg Stein <gs...@gmail.com> wrote:
> On Mon, Mar 19, 2012 at 16:48, Hyrum K Wright <hy...@wandisco.com> wrote:
>> [ warning: investigation is still ongoing, but I thought I'd report this here.]
>>
>> I'm trying to debug the Ev2 shims over ra_dav.  In doing so, I've
>> discovered an inconsistency between ra_serf and ra_neon (surprise!)
>
> ra_neon uses a "replay-report" while ra_serf uses an "update-report".
> The former provides the rev= attribute.
>
> I don't know why Neon uses the replay-report instead of the
> update-report. I dunno what the differences are.

What operations are you using? And why is it different between the two RA runs?

Note: in neon/fetch.c:1845, I see that SVN_INVALID_REVNUM is passed to
delete_entry(). That happens when ra_neon runs an update-report.

Cheers,
-g

Re: ra_serf not providing a valid revision to delete_entry() upon update

Posted by Greg Stein <gs...@gmail.com>.
On Mon, Mar 19, 2012 at 16:48, Hyrum K Wright <hy...@wandisco.com> wrote:
> [ warning: investigation is still ongoing, but I thought I'd report this here.]
>
> I'm trying to debug the Ev2 shims over ra_dav.  In doing so, I've
> discovered an inconsistency between ra_serf and ra_neon (surprise!)

ra_neon uses a "replay-report" while ra_serf uses an "update-report".
The former provides the rev= attribute.

I don't know why Neon uses the replay-report instead of the
update-report. I dunno what the differences are.

>...

Cheers,
-g

Re: ra_serf not providing a valid revision to delete_entry() upon update

Posted by Hyrum K Wright <hy...@wandisco.com>.
On Mon, Mar 19, 2012 at 3:48 PM, Hyrum K Wright
<hy...@wandisco.com> wrote:
> [ warning: investigation is still ongoing, but I thought I'd report this here.]
>
> I'm trying to debug the Ev2 shims over ra_dav.  In doing so, I've
> discovered an inconsistency between ra_serf and ra_neon (surprise!)
>
> ra_neon provides a valid argument for the replaced_rev parameter of
> the Ev1 delete_entry() function.  The relevant code is in
> libsvn_ra_neon/replay.c:
>
> 169         const char *path = svn_xml_get_attr_value("name", atts);
> 170         const char *crev = svn_xml_get_attr_value("rev", atts);
> 171
> 172         if (! path)
> 173           return MISSING_ATTR(nspace, elt_name, "name");
> 174         else if (! crev)
> 175           return MISSING_ATTR(nspace, elt_name, "rev");
> 176         else
> 177           {
> 178             dir_item_t *di = &TOP_DIR(rb);
> 179
> 180             SVN_ERR(rb->editor->delete_entry(path, SVN_STR_TO_REV(crev),
> 181                                              di->baton, di->pool));
> 182           }
> 183       }
> 184       break;
>
> Apparently, ra_neon expects and finds a "rev" attribute on the
> relevant XML element.  However, the same code in ra_serf (update.c)
> doesn't:
>
> 1556   else if ((state == OPEN_DIR || state == ADD_DIR) &&
> 1557            strcmp(name.name, "delete-entry") == 0)
> 1558     {
> 1559       const char *file_name;
> 1560       report_info_t *info;
> 1561       apr_pool_t *tmppool;
> 1562       const char *full_path;
> 1563
> 1564       file_name = svn_xml_get_attr_value("name", attrs);
> 1565
> 1566       if (!file_name)
> 1567         {
> 1568           return svn_error_create(
> 1569             SVN_ERR_RA_DAV_MALFORMED_DATA, NULL,
> 1570             _("Missing name attr in delete-entry element"));
> 1571         }
> 1572
> 1573       info = parser->state->private;
> 1574
> 1575       SVN_ERR(open_dir(info->dir));
> 1576
> 1577       tmppool = svn_pool_create(info->dir->dir_baton_pool);
> 1578
> 1579       full_path = svn_relpath_join(info->dir->name, file_name, tmppool);
> 1580
> 1581       SVN_ERR(info->dir->update_editor->delete_entry(full_path,
> 1582                                                      SVN_INVALID_REVNUM,
> 1583                                                      info->dir->dir_baton,
> 1584                                                      tmppool));
> 1585
> 1586       svn_pool_destroy(tmppool);
> 1587     }
>
> As you can see, ra_serf is unconditionally passing SVN_INVALID_REVNUM
> to delete_entry(), which is what is causing the shims to fail over
> ra_dav (when run with ra_serf).
>
> I tried inserting analogous code to fetch the "rev" attribute in
> ra_serf, but it appears that said attribute isn't available.  I'm not
> familiar enough with the protocol to know why it would be for ra_neon,
> but not ra_serf, and I'm asking here in the hopes that somebody can
> enlighten me. :)

For the record, the patch I tried was:

[[[
Index: subversion/libsvn_ra_serf/update.c
===================================================================
--- subversion/libsvn_ra_serf/update.c	(revision 1302583)
+++ subversion/libsvn_ra_serf/update.c	(working copy)
@@ -1557,11 +1557,13 @@
            strcmp(name.name, "delete-entry") == 0)
     {
       const char *file_name;
+      const char *crev;
       report_info_t *info;
       apr_pool_t *tmppool;
       const char *full_path;

       file_name = svn_xml_get_attr_value("name", attrs);
+      crev = svn_xml_get_attr_value("rev", attrs);

       if (!file_name)
         {
@@ -1579,7 +1581,7 @@
       full_path = svn_relpath_join(info->dir->name, file_name, tmppool);

       SVN_ERR(info->dir->update_editor->delete_entry(full_path,
-                                                     SVN_INVALID_REVNUM,
+                                                     SVN_STR_TO_REV(crev),
                                                      info->dir->dir_baton,
                                                      tmppool));
 ]]]

-Hyrum


-- 

uberSVN: Apache Subversion Made Easy
http://www.uberSVN.com/