Support more geospatial files.

[manthey]

There are files that have coordinates in lat/lon but no listed projection.

Closes #676.

[banesullivan]

This still isn't working for the specific data that I have

[banesullivan]

I'm seeing segfaults....

(large_image_dev) ➜  large_image git:(more-geospatial) python
Python 3.8.12 | packaged by conda-forge | (default, Sep 16 2021, 01:59:00)
[Clang 11.1.0 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import large_image
>>> large_image.__file__
'/Users/bane/Software/kw/large_image/large_image/__init__.py'
>>> large_image.open('/Users/bane/Software/OASIS/data/og_Data_scripts_Sept2021/PRISM_PPT/2017/PRISM_PPT_20170714.tif')
Created LRU Cache for 'tilesource' with 81 maximum size
TIFFReadDirectory: Warning, Unknown field with tag 42113 (0xa481) encountered.
[1]    7133 segmentation fault  python

[manthey]

This still isn't working for the specific data that I have

It works for the sample file you shared privately. There is a structurally similar file here: https://data.kitware.com/api/v1/file/61969eb32fa25629b9f3638b/download

[banesullivan]

I'm getting segfaults for that file as well so something is probably wrong with my local installation. I'll investigate this soon here

[manthey]

I'm seeing segfaults....

(large_image_dev) ➜  large_image git:(more-geospatial) python
Python 3.8.12 | packaged by conda-forge | (default, Sep 16 2021, 01:59:00)
[Clang 11.1.0 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import large_image
>>> large_image.__file__
'/Users/bane/Software/kw/large_image/large_image/__init__.py'
>>> large_image.open('/Users/bane/Software/OASIS/data/og_Data_scripts_Sept2021/PRISM_PPT/2017/PRISM_PPT_20170714.tif')
Created LRU Cache for 'tilesource' with 81 maximum size
TIFFReadDirectory: Warning, Unknown field with tag 42113 (0xa481) encountered.
[1]    7133 segmentation fault  python

Based on that segfault, it isn't trying to parse it as geospatial -- it is using some other tile source. What do you see with something like

import large_image

print(large_image.tilesource.isGeospatial(<path>))

[banesullivan]

Same segfault

>>> import large_image
>>> p = '/Users/bane/Software/OASIS/data/og_Data_scripts_Sept2021/PRISM_PPT/2017/PRISM_PPT_20170714.tif'
>>> large_image.tilesource.isGeospatial(p)
TIFFReadDirectory: Warning, Unknown field with tag 42113 (0xa481) encountered.
[1]    13148 segmentation fault  python

[manthey]

Same segfault

>>> import large_image
>>> p = '/Users/bane/Software/OASIS/data/og_Data_scripts_Sept2021/PRISM_PPT/2017/PRISM_PPT_20170714.tif'
>>> large_image.tilesource.isGeospatial(p)
TIFFReadDirectory: Warning, Unknown field with tag 42113 (0xa481) encountered.
[1]    13148 segmentation fault  python

Odd -- what version of the GDAL wheel do you have installed?

[banesullivan]

I was on 3.3.2 and so I reinstalled to 3.4.0 it's working now... 🤷🏻‍♂️

[manthey]

I was on 3.3.2 and so I reinstalled to 3.4.0 it's working now... 🤷🏻‍♂️

Were you importing any other libraries before large_image (this may or may not be germane, since the problem might occur even without such an import)? There was a curious issue with importing torch before gdal, which turned out that one of libraries that we were building for GDAL happened to use -ftls-model=initial-exec which exhausted some local resource for static thread local storage which caused a failed import. In other bug reports on this topic it causes segfaults.

[banesullivan]

Were you importing any other libraries before large_image

Nope, just large_image. I think I had a mismatch of versions between my python gdal bindings and the gdal binaries (on Mac so I can't use the large_image_wheels)