[mapserver-dev] RFC: GMaps API for mapserv

[Stephen Woodbridge]

Paul Spencer wrote:
> 
> On 10-Apr-08, at 5:33 PM, Stephen Woodbridge wrote:
>> Paul Ramsey wrote:
>>> On Thu, Apr 10, 2008 at 1:07 PM, Stephen Woodbridge
>>> <woodbri at swoodbridge.com> wrote:
>>>> It seems that the meta-tiling and labeling issues will need to be 
>>>> addressed
>>>> eventually, so I think some discussion of how they would be done from a
>>>> design point of view should be undertaken. If these can not be 
>>>> easily solved
>>>> in this architecture then that might indicate a problem.
>>> Metatiling implies caching, since you'll necessarily be producing way
>>> more tiles for each request than you need, you should store the spares
>>> for the future. And I don't think we want to go there. At least, some
>>> folks don't want to go there.
>>
>> Yes, I agree with the fact the metatiling is out of scope. I think the 
>> point that I did not make so clear is that assuming someone wants to 
>> do metatiling, say using some script or tilecache or whatever. Is 
>> there anything in the implementation or design that would prevent one 
>> from generating metatiles and chopping them up via a script and 
>> storing them somewhere outside of mapserver? If yes, is removing that 
>> within the scope of things or not?
> 
> I think that you should do a couple of things differently in mode=gmap:
> 
> * Steve Lime just added a processing option to enable mapserver to 
> calculate label positions before features are clipped.  With PARTIALS 
> TRUE, this should eliminate the label problems to a large extent when 
> producing tiles since adjacent tiles should put the label in the same 
> place (relatively speaking) and labels should end up continuing nicely 
> across tile boundaries.  So this processing option and PARTIALS TRUE 
> could be forced on in this specific mode
> 
> * draw map tiles a bit bigger and trim the result to reduce edge effects.
> 
> If you did this, with an apache module, meta tiles would be a thing of 
> the past - well maybe not, but it would go a long way to avoiding the 
> need for them.

Yes, but for performance reasons 5x5 metatiles require you to make one 
pass over the data on disk instead of 25 passes of the same data and 
even with file caching this is a significant performance boost if you 
have to generate a few billion tiles.

-Steve W

> Also, I think you should call it mode=tile :)
> 
> And ... +1!
> 
> Cheers
> 
> Paul
> __________________________________________
> 
>    Paul Spencer
>    Chief Technology Officer
>    DM Solutions Group Inc
>    http://www.dmsolutions.ca/
>