[Community] Shapely, ctypes and finding geos_c

[Sean Gillies]

Sean Gillies wrote:
> Christopher Barker wrote:
>> Hi all,
>>
>> I started checking out Shapely. Of course, the first step is getting it 
>> running!
>>
>> I'm running on OS-X 10.4. I've got the geos libs in a Framework, 
>> provided by KyngChaos:
>>
>> http://www.kyngchaos.com/wiki/software:frameworks
>>
>> William has done a great job of bundling up all sort of Open Source GIS 
>> software in Frameworks, which is, nominally, the "right way" to do it on 
>> OS-X.
>>
>> However, ctype.util.find_library doesn't find it.
>>
>> This is a general issue with ctypes-based stuff. I can imagine this is 
>> always a pain, unless you deliver the libs with your own code. In any 
>> case, barring digging into the find_library code and trying to patch 
>> that, I added a lit bit to shapely.geos:
>>
>> elif sys.platform == 'darwin':
>>      lib = find_library('geos_c')
>>      if lib is None:
>>          ## try a few more locations
>>          ## This, or course, could be a lot smarter!
>>          lib_paths = 
>> ["/Library/Frameworks/GEOS.framework/Versions/Current/unix/lib/libgeos_c.dylib",# 
>> The Framework build from Kyng Chaos:
>>          ]
>>          for path in lib_paths:
>>              if os.path.exists(path):
>>                  lib = path
>>                  break
>>          else:
>>              raise Exception("Could not find geos_c library")
>>      lgeos = CDLL(lib)
>>      free = CDLL(find_library('libc')).free
>>
>> This works for me, and allows future folks to add other likely paths to 
>> that list (darwinports, fink, ???)
>>
>> Is this worth a patch?
>>
>>
>> Also, I was poking around a bit, and did this:
>>
>> (poly is a Polygon object)
>>
>> cs = poly.get_coords()
>>  >>> cs
>> <shapely.geometry.base.CoordinateSequence object at 0x71e30>
>>  >>> import numpy as N
>>  >>> N.asarray(cs)
>> Bus error
>>
>> oops!
>>
>> Maybe I shouldn't be able to turn a coords object into a numpy array, 
>> but a crash should never happen.
>>
>> -Chris
> 
> Well, you were as welcome as anyone to beta test it so the bug gets
> caught before 1.0 ;) Maybe Shapely coordinate sequences *should* provide
> the Numpy array interface after all. I can't think of a reason why not.
> 
> The only problem I see with the patch is that the framework name/path
> looks arbitrary to me. But then I don't know much about the Mac lib
> frameworks.
> 
> Cheers,
> Sean

New tickets:

http://trac.gispython.org/projects/PCL/ticket/152
http://trac.gispython.org/projects/PCL/ticket/153

Sean