IPublishTraverse that is part of a view not being invoked

[Martin Aspeli]

I believe (but I'm unable to write a test now) that if a view implements IPublishTraverse, its publishTraverse() method is never called.

To check, try to edit adding.py in Five and put a pdb or print statement in publishTraverse(). Go to /foo/bar/+/foobar - if I understand things correctly, this should invoke publishTraverse(name='foobar'), but in my tests it does not.

[Philipp von Weitershausen (philikon)]

I think what Martin means is that given a URL path like a/b/c with the 'b' object providing IPublishTraverse, the ZPublisher will still look up an IPublishTraverse multi-adapter and not take advantage of b already providing the IPublishTraverse API. 'b' can also be a view, but in general, it can be any object.

[Lennart Regebro (regebro-gmail)]

No, the ZPublisher will check if the object provides IPublishTraverse and use it directly. I tested this with a
view and it works fine.

[Lennart Regebro (regebro-gmail)]

Changes: submitter email, edited transcript

[Martin Aspeli (optilude)]

So, would you say Five/browser/adding.py is an example of something that ought to work? I can't get its publishTraverse() to invoke if I e.g. do

http://localhost:8080/site/folder/+/foobar

putting a pdb in publishTraverse() should cause a debugger session and give me name='foobar', should it not? It didn't for me :-/

Martin

[Bertrand Mathieu (bmat)]

Got it with zope 2.10.5. The culprit is Five metaconfigure.I have made a patch with tests, but the error seems obvious: just compare Products/Five/browser/metaconfigure.py and zope/app/publisher/browser/viewmeta.py. The code in Five seems to be an old copy/paste.

In short: the view class is wrapped by Five. If it has a 'publishTraverse' method it should be called by the wrapper. But the code get the method with: class_.publishTraverse.__get__(self).__of__(self) => __of__ on a method?! The code in zope.app is the same without __of__
I guess there is a try/except somewhere that hides this error.

The patch is made against svn://svn.zope.org/repos/main/Products.Five/branches/1.5 at rev80401.

[Philipp von Weitershausen (philikon)]

On 18 Apr 2008, at 14:11 , Bertrand Mathieu wrote:
> Got it with zope 2.10.5. The culprit is Five metaconfigure.I have
> made a
> patch with tests, but the error seems obvious: just compare
> Products/Five/browser/metaconfigure.py and
> zope/app/publisher/browser/viewmeta.py. The code in Five seems to be
> an
> old copy/paste.
>
> In short: the view class is wrapped by Five. If it has a
> 'publishTraverse' method it should be called by the wrapper. But the
> code get the method with:
> class_.publishTraverse.__get__(self).__of__(self) => __of__ on a
> method?! The code in zope.app is the same without __of__
> I guess there is a try/except somewhere that hides this error.

This code is indeed strange. Your observations seem to be correct.
However, it would be good to see how you tested this...

> The patch is made against
> svn://svn.zope.org/repos/main/Products.Five/branches/1.5 at rev80401.

Thanks for the patch. It seems to be missing the 'view.txt' file,
though, which contains the test. This is the interesting bit :).

[Bertrand Mathieu (bmat)]

oh too bad. Sorry! I quickly made a svn diff, but forgot svn add view.txt... I'll see if I can rewrite view.txt, I've lost it :-/.

Basically I don't know how to call the publish traverse process, so for my test I get the view instance from the test content, and call directly view.publishTraverse('traversed') on it; because the actual bug is that Five "publishTraverse wrapper" fails at finally calling the method, I think it is correct.

[Philipp von Weitershausen (philikon)]

On 18 Apr 2008, at 14:42 , Bertrand Mathieu wrote:
> oh too bad. Sorry! I quickly made a svn diff, but forgot svn add
> view.txt... I'll see if I can rewrite view.txt, I've lost it :-/.

Argh, why would you delete it? Anyway, perhaps it's enough just to add
another paragraph/section to an existing test?

> Basically I don't know how to call the publish traverse process, so
> for
> my test I get the view instance from the test content, and call
> directly
> view.publishTraverse('traversed') on it; because the actual bug is
> that
> Five "publishTraverse wrapper" fails at finally calling the method, I
> think it is correct.

Please look at the other tests (e.g. pages.txt, pages_ftest.txt)!

[Jens Vagelpohl (dataflake-deactivatedaccount-deactivatedaccount)]

The supposed bad wrapping using "__of__" disappeared shortly after this bug was filed, closing as resolved:

http://svn.zope.org/?rev=85767&view=rev