Howard Butler wrote:
> On Aug 22, 2008, at 4:10 PM, Sean Gillies wrote:
> 
>> Jaakko Salli wrote:
>>> Sean Gillies wrote:
>>>> Hi all,
>>>>
>>>> If I remember correctly, the blocker for Rtree 1.0
>>>>
>>>>  http://trac.gispython.org/lab/milestone/Rtree%201.0
>>>>
>>>> is a unwritten want to have the Spatialindex library build on  
>>>> Windows.
>>>>
>>> What problems exactly did you have building spatialindex on Windows?
>>>
>>> Regards,
>>>  Jaakko
>> Hi Jaakko,
>>
>> I haven't had any problems, myself, but then I'm not a Windows user ;)
>> Do you have a DLL that works? If so, could I persuade you to make an
>> Rtree bdist_winist [1] for us?
>>
>> Back to the 1.0 idea here ... Howard expressed a desire to switch  
>> Rtree
>> over to using ctypes before 1.0. I'd rather just get it out there as  
>> it
>> is since it's functionally complete, and perhaps switch to ctypes in a
>> future milestone. Howard, how about we resolve this as you do in
>> MapServer? That is, a -1 vote on pushing Rtree forward to 1.0 as it is
>> carries with it the obligation to take the lead on a ctypes  
>> implementation.
> 
> 
> I won't veto it if a veto takes on that responsibility.  I don't think  
> the current Rtree should be called 1.0, but I'm not too picky on  
> names.  Here's some of the reasons why I don't think we're at 1.0 yet:
> 
> - We don't work on windows for file-based index storage (this is  
> spatialindex's fault of course).

It doesn't? I don't see any mention of that at

http://trac.gispython.org/spatialindex/query?status=new&status=assigned&status=reopened&status=closed&order=priority

I did find a mention here

http://lists.gispython.org/pipermail/spatialindex/2008-January/000012.html

I'm not sure who will *ever* be motivated to fix this.

> - We only support an extremely simple subset of what spatialindex  
> supports, with very little configuration for a user who knows their  
> way around r*trees.  Performance/memory issues that we might see from  
> Python can easily be blamed on our limited configuration ability  
> rather than the spatialindex library.

Sure, but I think what we do provide is good enough for a 1.0.

> - The C/C++ interface between spatialindex and the Python Rtree code  
> is less than ideal, and I don't think it is sustainable for long term  
> development if the goal above is also a goal of the project.
> 

Maybe. I suspect that there will never be many people motivated to do
anything about configuration, C or ctypes. I think it's you and me no
matter what. For example: Shapely uses ctypes, but I can't find more
than 1 instance of a patch involving ctypes. We've already received that
many patches for Rtree's C/C++ interface. Within these projects there is
no demonstrated win (yet) for ctypes.

> I do desire this stuff to be improved, but I have no time/money/client/ 
> business case to drive it forward.  If we want to slap a 1.0 on what  
> we have and call the next 2.0, that's fine with me too.  Whatever we  
> call it, it's not *done* in my mind.
> 
> Howard

I don't think it's done either. But it is satisfying its main customer,
zgeo.spatialindex, and meeting my initial design goals.

Sean

_______________________________________________
Community mailing list
[email protected]
http://lists.gispython.org/mailman/listinfo/community

Reply via email to