feedback on possible mapserver enhancements

[Arnulf Christl]

Pericles S. Nacionales wrote:
> Howard Butler wrote:
>> On Feb 4, 2008, at 11:15 AM, Daniel Morissette wrote:
>>
>>> thomas bonfort wrote:
>>>>>> And last but not least :
>>>>>> * what would you think of having a wfs-t implementation for 
>>>>>> mapserver,
>>>>>> probably at first limited to postgis backends, and based on the
>>>>>> tinyows project?
>>>>> A year ago I would have said no, but several times in recent months 
>>>>> I've had questions
>>>>> from folks that seem to use WFS-T as a means of selecting their web 
>>>>> rendering tool. It's
>>>>> becoming a differentiating feature. I'm not familiar with TinyOWS 
>>>>> though. Are you
>>>>> suggesting assimilating TinyOWS?
>>>> the advantage of this would be to avoid having to deploy another
>>>> server along side mapserver in order to treat the wfs-t side of an
>>>> application,as you pointed out. in finality it would mean porting of
>>>> the tinyows code into mapserver.
>>>
>>> There is so much demand for WFS-T by our users that I am slowly 
>>> giving up and starting to think that we may have to do WFS-T in the 
>>> end. Please don't tell anyone that I wrote that. ;) ;)
>>>
>>> I am not sure about integrating TinyOWS code... I have never looked 
>>> at TinyOWS, but wouldn't a simple merge be messy? How would that fit 
>>> with existing mapwfs.c code? Could we not just extend the current 
>>> implementation (and make the necessary architecture changes) to 
>>> support transactions?
>>
>>
>> MapServer is not a GIS!  MapServer is not a GIS!  I am not supportive 
>> at all of implementing WFS-T in MapServer.  What benefit is there to 
>> be gained by doing so that can't be accomplished by setting up a 
>> GeoServer instance alongside MapServer?  IMO, it is the best-of-breed 
>> open source WFS-T that's out there, with tons of momentum and 
>> development force behind it -- why go to the trouble to re-implement 
>> it in MapServer?
>>
>> Technically, one challenge I see for MapServer implementing WFS-T is 
>> that MapServer apps generally expect to be transient and stateless.  
>> MapServer does not do well in long running processes (any MapScripter 
>> who's tried can give you gobs of complaints about this), and it has no 
>> concept of transactional operations which I think would be very 
>> challenging to bolt on in any smooth sort of way.
>>
>> IMO, MapServer should continue to improve upon what it is good at, and 
>> WFS-T is not something that I think it would be good at without a lot 
>> of re-engineering (we hate churn, remember?).  With some effort, we 
>> could have something workable and maybe even functional, but it will 
>> get nowhere close to what GeoServer has.
> My $0.02...  I agree that MapServer is not a GIS.  But adding a WFS-T 
> support still doesn't make MapServer a GIS.  ...maybe one step closer. 
> :)  Also, what's the point for a user to install GeoServer and MapServer 
> when GeoServer can do what MapServer does?  Won't you agree that 
> MapServer loses out in that scenario?  I doubt many users would want to 
> install both and manage all those configuration files.  I also agree 
> that it would be a technical challenge to add WFS-T support in MapServer 
> but having TinyOWS code already available for review/merge should help 
> some in that regard.
> 
> If WFS-T is to be supported--and I support the idea--it should be 
> optional and that it should not have a drastic effect on MapServer's 
> rendering performance.
> 
> -Perry

Hello,
if I may interject... Why don't you include a simple transactional RESTful WFS? That way you can stay stateless which was one of the major issues iirc. I think Sean Gillies or Chris Schmidt were doing some experiments (I can look up the links if it is of interest). This would give people a quick alternative to get going. I doubt that we really need a full fledged WFS-T with all the related problems. 

Having MapServer and GeoServer in one infrastructure is good for us consultants and is appropriate for large infrastructures running on several distributed machines. But it is a pain to have to maintain a CGI interface and a servlet container and to plunge into a full fledged GeoServer if all you want is to do is write back a few point coordinates. (/me runs for cover...)

Best regards, 
Arnulf.