[Zope-dev] zope.publisher 3.5 branch has code/behavior not a part of subsequent releases

[Gary Poster]

On Aug 24, 2009, at 5:27 PM, Tres Seaver wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Gary Poster wrote:
>> Hm.  I sent this from the wrong account, so it didn't make it to the
>> zope-dev list.  I'm also adding an additional bit of war story at the
>> end.
>>
>>
>> On Aug 24, 2009, at 11:16 AM, Gary Poster wrote:
>>
>>> Hi Tres
>>>
>>> I made a 3.5.8 release of the zope.publisher 3.5 branch for a reason
>>> unimportant to this email (I backported a change to trunk that was
>>> discussed on the mailing list).  It looks like you made a 3.5.7 with
>>> the following change (omitting tests and the like).
>>>
>>> 98932    tseaver                 # Python 2.6 notices QS-on-POST,
>>> which breaks BBB for us.
>>> 98932    tseaver                 # Suppress that.
>>> 98932    tseaver                 if 'QUERY_STRING' in self._environ:
>>> 98932    tseaver                     env = self._environ
>>> 98932    tseaver                     env['X-POST-QUERY_STRING'] =
>>> env.pop('QUERY_STRING')
>>>
>>> I can handle adjusting to this change, if it is appropriate, but my
>>> concern is that it is not in trunk or any subsequent major release
>>> (3.6, 3.7, 3.8) of zope.publisher.  Therefore, if I change my code
>>> to use my updated 3.5 release, which I had hoped would be a
>>> conservative update, I have to change in a currently forward-
>>> incompatible way.
>>>
>>> What's the story here?  Is 3.5.7 a brownbag that needs to have its
>>> changes aborted in a subsequent release in that branch?  Or are
>>> those legitimate changes that need to be forward ported?
>>>
>>> FWIW, we (Launchpad) also care about this case of a POST with other
>>> pertinent, separate data in the query string, and the behavior you
>>> implement here would be fine if it is necessary for Py 2.6 as your
>>> comment describes.
>>
>> Also FWIW, this change breaks many of our forms that were explicitly
>> constructing form actions that included the current query string.   
>> I'm
>> not sure how that could be avoided, although the fix might have been
>> simpler if there were always an X-ORIGINAL-QUERY_STRING or something
>> else.
>>
>> If I were not already behind, I would investigate to understand the
>> Python 2.6 problem better and see what other frameworks are doing
>> here.  I understand from conversations with other engineers that at
>> least some Django developers are accustomed to always having access  
>> to
>> the query string on the request object, whether the method were get  
>> or
>> post or whatever else.
>
> It is *essential* for correct operation that QUERY_STRING values *not*
> be admixed with POSTed form values.  I don't really care how we  
> resolve
> your issue, as long as we do not end up in a case where the values in
> the query string get mingled into the form data:  for instance, we  
> could
> hand a QUERY_STRING-free copy of the environment to the  
> cgi.FieldStorage
> machinery.
>
> Whatever gets done needs to leave the existing test in place::
>
>   self.assertEqual(dict(request.form), dict(x='1', y='2'))
>
> for a request whose QUERY_STRING was 'a=5&b:int=6', but which posted  
> the
> 'x' and 'y' values.

I'm good with that.

My additional constraint would be that

self.assertEqual(request.get('QUERY_STRING'), 'a=5&b:int=6')

for the same request.  Hiding the QUERY_STRING causes us some  
unpleasant and subtle functional test pain.

Passing a QUERY_STRING-free copy to cgi.FieldStorage as you suggest  
seems like a good way to go to me.

Unless someone beats me to it or stops me, I'll plan to make this go  
in trunk and in the 3.5 branch, and make a 3.5.9 release.

(If someone *were* to beat me to it, that would be awesome.  I'm kinda  
swamped.)

Thanks

Gary