[Phoenix] Parser exception re introspection files

[Andreas Wuest]

Hi

On 28.9.2006 14:52 Uhr, Michael Wechner wrote:

> Andreas Wuest wrote:
> 
>> Hi
>>
>> On 28.9.2006 10:14 Uhr, Michael Wechner wrote:
>>
>>> Hi
>>>
>>> I just realized that Yulup rc5 throws an exception when trying to parse
>>>
>>>
>>> <navigation>
>>>  <sitetree href="sitetree.xml" method="GET"/>
>>> </navigation>
>>>
>>> within an introspection file. Gregor told that this is because Yulup 
>>> rc5  has this "method" kind of implemented, but
>>> not fully. I have commented it for the moment such that the demo is 
>>> working again for people using Yulup rc5.
>>>
>>> But how can we prevent that future versions have similar problems?
>>>
>>> Shouldn't we upgrade the version within the namespace when we 
>>> introduce new elements, such that Yulup can check and doesn't get a 
>>> hickup?
>>
>>
>> Sure, that would be the cleanest solution, although it somehow kills 
>> the idea of forward-compatible... ;)
> 
> that shouldn't be of course

That's why I think we need a strategy for changing version numbers. On 
the one hand, the specification version should clearly indicate that 
something changed, i.e. an element was added. OTOH, the namespace of the 
introspection document should not change in such a situation, because 
new elements don't do harm. So there is not a direct relationship 
between the introspection document version and the actual specification 
version.

>> If Yulup would run into an introspection document with a higher 
>> Neutron version than it supports, it would simply say "don't know 
>> about this, can't use", and halt.
>>
>> Anyway, when introcuding *new* elements, they should simply be ignored 
>> by a client,
> 
> exactly
> 
>> and therefore there would be no need to force a client to fail by 
>> bumping the Neutron version number.
>>
>> See e.g. this document http://www.w3.org/2001/tag/doc/versioning for a 
>> discussion of versioning issues and how to resolve them with regard to 
>> XML languages.
> 
> I guess for the moment we should write down a test-suite which the 
> release manager needs to check before actually releasing.
> 
> I will start something ;-)

Great! :) See bug 
http://bugzilla.wyona.com/cgi-bin/bugzilla/show_bug.cgi?id=4636 which 
I've opened about 10 years ago...

In fact, the RM should do several things: she/he should run the test 
suite. But she/he must also make sure that none of the developers 
currently has code checked in which is not yet finished, i.e. which may 
break.

-- 
Kind regards,
Andi