On Thu, May 20, 2010 at 4:59 PM, Greg Stein <gstein_at_gmail.com> wrote:
> Thanks for the ping.
>
> The patch looks good except for the incoming-delete case.
Hi Greg,
Which flavor of that case? Incoming delete on a local delete of the
same property with the same value? Or something else?
> If the
> svn_string_compare() succeeds, but mine==NULL, then you get the crash.
> I think the mine==NULL needs to remain on the outer-if test.
I'm not entirely sure what you are referring to here. Is it this
section of generate_conflict_message()?
[[[
if (svn_string_compare(original, incoming_base))
{
if (mine)
/* We were trying to delete the correct property, but an edit
caused the conflict. */
return svn_string_createf(result_pool,
_("Trying to delete property '%s' with "
"value '%s',\nbut it has been modified "
"from '%s' to '%s'."),
propname, incoming_base->data,
original->data, mine->data);
}
else if (mine == NULL)
{
/* We were trying to delete the property, but we have locally
deleted the same property, but with a different value. */
return svn_string_createf(result_pool,
_("Trying to delete property '%s' with "
"value '%s',\nbut property with value "
"'%s' is locally deleted."),
propname, incoming_base->data,
original->data);
}
/* We were trying to delete INCOMING_BASE but our ORIGINAL is
something else entirely. */
SVN_ERR_ASSERT_NO_RETURN(!svn_string_compare(original, incoming_base));
]]]
If (ORIGINAL == INCOMING_BASE) && (MINE == INCOMING == NULL) then
we'll trigger that SVN_ERR_ASSERT_NO_RETURN. But we shouldn't be
calling this function in the first place for this case, because the
function assumes there *is* a prop conflict of some kind. It always
produces a conflict message or asserts trying.
At any rate, I'm a bit confused here.
Paul
> Other than that... looks great. Commit!
>
> Cheers,
> -g
>
> On Thu, May 20, 2010 at 15:26, Paul Burba <ptburba_at_gmail.com> wrote:
>> Hi Greg,
>>
>> If you have a chance, let me know if you were planning on giving any
>> feedback on this. Just want to be sure I answered your questions
>> before committing.
>>
>> Thanks,
>>
>> Paul
>>
>> On Fri, May 14, 2010 at 1:46 PM, Paul Burba <ptburba_at_gmail.com> wrote:
>>> On Fri, Apr 23, 2010 at 5:22 PM, <gstein_at_apache.org> wrote:
>>>> Author: gstein
>>>> Date: Fri Apr 23 21:22:52 2010
>>>> New Revision: 937524
>>>>
>>>> URL: http://svn.apache.org/viewvc?rev=937524&view=rev
>>>> Log:
>>>> Begin new infrastructure for generating prop conflict messages. This will
>>>> allow us to (re)generate a property reject file at will, given a record of
>>>> the property conflicts on a given node.
>>>>
>>>> There are two issues for discussion and fixing in a future revision:
>>>> - incoming-delete will remove local-add (it should conflict?)
>>>
>>> Hi Greg,
>>>
>>> I think the correct behavior is: An incoming-delete removes a local
>>> add only if the incoming base value is the *same* as the added value;
>>> otherwise there is a conflict. This is analogous to how we treat an
>>> incoming file deletion on a local file addition. It's only a tree
>>> conflict if the files differ.
>>>
>>> More below...
>>>
>>>> - incoming-delete will crash on a local-delete
>>>>
>>>> * subversion/libsvn_wc/props.c:
>>>> (generate_conflict_message): new function to generate a property
>>>> conflict message given the four property values involved in a 4-way
>>>> merge.
>>>> (apply_single_prop_delete): leave two notes about behavior in here (see
>>>> the issues above). fix message generation: use OLD_VAL, not BASE_VAL
>>>> (apply_single_generic_prop_change): the OLD_VAL parameter will always be
>>>> not-NULL, so we can simplify an if condition.
>>>> (svn_wc__merge_props): save away MINE_VAL, and then if we see a conflict
>>>> message returned by the property merging functions, then assert that
>>>> our new function comes up with the same message
>>>>
>>>> * subversion/tests/cmdline/prop_tests.py:
>>>> (prop_reject_grind): new test function to grind thru all the variations
>>>> of property conflicts.
>>>> (test_list): add new test
>>>>
>>>> * subversion/tests/cmdline/svntest/sandbox.py:
>>>> (Sandbox.simple_propset, Sandbox.simple_propdel): new methods
>>>>
>>>> Modified:
>>>> subversion/trunk/subversion/libsvn_wc/props.c
>>>> subversion/trunk/subversion/tests/cmdline/prop_tests.py
>>>> subversion/trunk/subversion/tests/cmdline/svntest/sandbox.py
>>>>
>>>> Modified: subversion/trunk/subversion/libsvn_wc/props.c
>>>> URL: http://svn.apache.org/viewvc/subversion/trunk/subversion/libsvn_wc/props.c?rev=937524&r1=937523&r2=937524&view=diff
>>>> ==============================================================================
>>>> --- subversion/trunk/subversion/libsvn_wc/props.c (original)
>>>> +++ subversion/trunk/subversion/libsvn_wc/props.c Fri Apr 23 21:22:52 2010
>>>> @@ -709,6 +709,136 @@ svn_wc_merge_props3(svn_wc_notify_state_
>>>> }
>>>>
>>>>
>>>> +/* Generate a message to describe the property conflict among these four
>>>> + values.
>>>> +
>>>> + Note that this function (currently) interprets the property values as
>>>> + strings, but they could actually be binary values. We'll keep the
>>>> + types as svn_string_t in case we fix this in the future. */
>>>> +static const svn_string_t *
>>>> +generate_conflict_message(const char *propname,
>>>> + const svn_string_t *original,
>>>> + const svn_string_t *mine,
>>>> + const svn_string_t *incoming,
>>>> + const svn_string_t *incoming_base,
>>>> + apr_pool_t *result_pool)
>>>> +{
>>>> + if (incoming_base == NULL)
>>>> + {
>>>> + /* Attempting to add the value INCOMING. */
>>>> + SVN_ERR_ASSERT_NO_RETURN(incoming != NULL);
>>>> +
>>>> + if (mine)
>>>> + {
>>>> + /* To have a conflict, these must be different. */
>>>> + SVN_ERR_ASSERT_NO_RETURN(!svn_string_compare(mine, incoming));
>>>> +
>>>> + /* Note that we don't care whether MINE is locally-added or
>>>> + edited, or just something different that is a copy of the
>>>> + pristine ORIGINAL. */
>>>> + return svn_string_createf(result_pool,
>>>> + _("Trying to add new property '%s' with "
>>>> + "value '%s',\nbut property already "
>>>> + "exists with value '%s'."),
>>>> + propname, incoming->data, mine->data);
>>>> + }
>>>> +
>>>> + /* To have a conflict, we must have an ORIGINAL which has been
>>>> + locally-deleted. */
>>>> + SVN_ERR_ASSERT_NO_RETURN(original != NULL);
>>>> + return svn_string_createf(result_pool,
>>>> + _("Trying to create property '%s' with "
>>>> + "value '%s',\nbut it has been locally "
>>>> + "deleted."),
>>>> + propname, incoming->data);
>>>> + }
>>>> +
>>>> + if (incoming == NULL)
>>>> + {
>>>> + /* Attempting to delete the value INCOMING_BASE. */
>>>> + SVN_ERR_ASSERT_NO_RETURN(incoming_base != NULL);
>>>> +
>>>> + /* A conflict can only occur if we originally had the property;
>>>> + otherwise, we would have merged the property-delete into the
>>>> + non-existent property. */
>>>> + SVN_ERR_ASSERT_NO_RETURN(original != NULL);
>>>> +
>>>> + if (mine && svn_string_compare(original, incoming_base))
>>>> + {
>>>> + /* We were trying to delete the correct property, but an edit
>>>> + caused the conflict. */
>>>> + return svn_string_createf(result_pool,
>>>> + _("Trying to delete property '%s' with "
>>>> + "value '%s'\nbut it has been modified "
>>>> + "from '%s' to '%s'."),
>>>> + propname, incoming_base->data,
>>>> + original->data, mine->data);
>>>> + }
>>>> +
>>>> + /* We were trying to delete INCOMING_BASE but our ORIGINAL is
>>>> + something else entirely. */
>>>> + SVN_ERR_ASSERT_NO_RETURN(!svn_string_compare(original, incoming_base));
>>>> +
>>>> + /* ### wait. what if we had a different property and locally
>>>> + ### deleted it? the statement below is gonna blow up.
>>>> + ### we could have: local-add, local-edit, local-del, or just
>>>> + ### something different (and unchanged). */
>>>
>>> <snip>
>>>
>>>> @@ -1166,6 +1296,8 @@ apply_single_prop_delete(svn_wc_notify_s
>>>>
>>>> if (! base_val)
>>>> {
>>>> + /* ### what about working_val? what if we locally-added? */
>>>> +
>>>> apr_hash_set(working_props, propname, APR_HASH_KEY_STRING, NULL);
>>>> if (old_val)
>>>> /* This is a merge, merging a delete into non-existent */
>>>> @@ -1216,11 +1348,13 @@ apply_single_prop_delete(svn_wc_notify_s
>>>> cancel_func, cancel_baton,
>>>> dry_run, scratch_pool));
>>>> if (got_conflict)
>>>> + /* ### wait. what if we had a different property and locally
>>>> + ### deleted it? the statement below is gonna blow up. */
>>>
>>> Attached is a patch that fixes the segfault and makes an incoming
>>> deletion on a local addition, where the incoming base value differs
>>> from the added value, a conflict, rather than unconditionally deleting
>>> the addition.
>>>
>>> I also tweaked prop_test.py 32 to check the results of the *.prej file.
>>>
>>> This patch adds two new potential conflicts messages:
>>>
>>> Incoming delete on local add of different value:
>>>
>>> Trying to delete property 'del.add' with value 'repos',
>>> but property has been locally added with value 'local'.
>>>
>>> Incoming delete on local delete of different value:
>>>
>>> Trying to delete property 'del.del' with value 'repos',
>>> but property with value 'local' is locally deleted.
>>>
>>> What do you think?
>>>
>>> Paul
>>>
>>> [[[
>>> Fix some property merge conflict bugs.
>>>
>>> 1) Incoming delete on a local add of a different value is now a
>>> conflict. Previously it was a clean merge and the prop was
>>> deleted.
>>>
>>> 2) Incoming delete on a local delete where the incoming base value
>>> differs from the local value is now a conflict. Previously
>>> this caused a segfault.
>>>
>>> * subversion/libsvn_wc/props.c
>>>
>>> (generate_conflict_message): Handle incoming delete on local add and
>>> incoming delete on local delete of a different prop value. Consistently
>>> use a trailing ',' after the first line of each prej conflict message.
>>>
>>> (apply_single_prop_delete): Stop considering an incoming delete on a local
>>> add as a merge.
>>>
>>> * subversion/tests/cmdline/prop_tests.py
>>>
>>> (prop_reject_grind): Start testing incoming delete on local delete of
>>> different prop value. Verify the resulting *.prej file.
>>> ]]]
>>>
>>
>
Received on 2010-05-21 16:54:20 CEST