Mark Phippard wrote:
> On Tue, Dec 9, 2008 at 3:45 PM, Hyrum K. Wright
> <hyrum_wright_at_mail.utexas.edu> wrote:
>> Mark Phippard wrote:
>>> On Sat, Dec 6, 2008 at 2:49 PM, Hyrum K. Wright
>>> <hyrum_wright_at_mail.utexas.edu> wrote:
>>>> Mark Phippard wrote:
>>>>> Just to follow-up, this problem still happens in trunk. I have
>>>>> attached a patch to the JavaHL tests that causes the breakage. I also
>>>>> looked into the NotifyInformation class and there is no way to fix
>>>>> this problem in that class as the error occurs in a base class from
>>>>> the JDK that it inherits from.
>>>>>
>>>>> I can commit the patch but it causes breakage. I thought it might be
>>>>> easier to just give to someone that is willing to look into the
>>>>> problem.
>>>> The attached patch fixes the problem locally by setting the commit item path to
>>>> the same value as the url. The result will be the url being passed as the path
>>>> parameter to the notification callback. Do we do that for the other items in a
>>>> wc-to-url copy, or are the notification paths some sort of relative path?
>>> It looks like you asked what my concerns were to this patch on IRC.
>>>
>>> My concern was mainly that I questioned whether it was valid to send a
>>> URL to that notification in the path field. It might be valid, the
>>> JavaDoc for the class does not really say. Given how I know we use
>>> this field in Subclipse, I believe we expect to receive local paths.
>>> That said, if SVN already sends URL's in other places, then we likely
>>> already have code to handle this.
>>>
>>> I asked if we needed to send this notification at all. It does not
>>> seem to be adding anything of value for an API consumer so why send
>>> it?
>> Well, I forget what we do on the command line for 'svn cp --parents $WCPATH
>> $URL', but I don't think we just ignore the parent notifications. We ought to
>> be doing the same thing in JavaHL which we are doing on the command line, IMHO.
>> (Or maybe the patch proves we're not doing the right thing on the commandline,
>> either. What do we do for URL->URL copies with parents?)
>
> It is possible the command line sees the null path and just doesn't
> pay attention. If the Java class had not been implemented the way it
> was, that is probably what I'd have done too. Just ignore it in Java.
> But since it uses that EventObject class from the JRE our hands are
> tied.
>
> I've not debugged URL-> URL but I'd guess there are no notifications.
> It looks like the notification we are getting the null path on in a WC
> -> URL copy is "commit_added" which seems odd. Is it for adding one
> of the parent paths?
Yes. When copying with 'cp --parents', the parents are considered added as part
of the commit.
> I guess if we just go with your code it would still be better.
> Because even if it trips up code like Subclipse at least it can be
> handled in Subclipse and not generate an exception in a Java base
> class.
We can, but if there's a more consistent (but still simple) way of solving the
problem, I'm happy to pursue that.
-Hyrum
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=981845
Received on 2008-12-09 21:59:16 CET