Error reading shapefile on Mac OS X with G5 processor

[Claire Mclister]

I've built mapnik on Mac OS X (Leopard) running on a G5 processor.

While trying to read the shapefile from the mappinghacks website (as suggested by the Getting Started page), I am getting an invalid envelope:

Shapefile(file='world_borders')

size = 2

file=world_borders

type=shape

Envelope(4.157455691574588e-317,-1.62334203800658e-118,4.157518931977256e-317,4.15549523909189e-317)

file_length=3332172

shape_type=5

datasource=0x332fd0 type=1

<mapnik.Datasource object at 0x64270>

[springmeyer]

Claire,

Try setting the shell env like this:

export DYLD_PRINT_APIS=y

before running the demo. Attach the full output of the python interpreter.

Also, confirm that you have tried running the demo.py as well or at least another world shapefile.

[Claire Mclister]

Output of running 'python rundemo.py'

[Claire Mclister]

I have attached the output from running the rundemo.py

Yes, confirmed, I have tried it with several shapefiles and for all the Envelope output doesn't look correct.

[springmeyer]

Hmmm. Nothing definitive in those logs (that I can see). I ran the demo on my machine, saved the same output and compared the text using BBedit. The most suspicious line (besides all the messed up extents) is line 355 where your log reports 'utf8' encoding then bails on processing the shapefile, while mine simply reports 'dlsym(0x32b7f0, create)'. Not sure what that means, but makes me wonder whether the some unicode reworking in Mapnik code is causing problems on Mac OS. I installed the boost ICU variant and ( I think) ICU separately as well using macports (can't recall exactly). Did you install ICU? Can you confirm that your libiconv is working as well?

My log is attached if you care to search through the haystack.

[springmeyer]

Output of running demo.py on intel 10.5 machine

[Claire Mclister]

I took a look at the logs, and was browsing the source code for reading shapefiles as well. Seems to me, the issue is being caused by an incorrect reading of doubles. Looks like integer values (file length) are being read properly, but doubles are getting messed up. Can anyone suggest a way to debug this? I'd be happy to modify the source code and try out different things.

[Claire Mclister]

It seems that the macro WORDS_BIGENDIAN is not being defined in Mac OS X with G5 processor, although it should. So, I manually defined it for the two files that it occurs in plugins/input/shape/shapefile.hpp and src/wkb.cpp and I can get the maps to work now.

Will try to figure out how to make sure the macro is defined in this case. Until then, will keep the hack on my copy.