Hi Sean

regarding the change in web mercator, I guess this is due to a subtle change where we don't check anymore if latitude is close enough of PI/2 to decide the point doesn't project. Now we only test lat < PI/2. And as 90° converted in radian is slightly < PI/2, it now projects to a finite (although very big) northing value. So this changes the aspect of how a raster with world wide bounds in EPSG:4326 reprojects in EPSG:3857. Neither the previous (which truncated a few pixels) nor the new behavior (which results in an image with a big height/width ratio) are ideal, but I don't think there isn't an ideal solution given the characteristics of the mercator projection. The commit that modified this is https://github.com/OSGeo/PROJ/commit/5d22d137b1be282bdc14c1d12ab8f669f58d41a6

Regarding discovery of proj.db, I can only think to https://github.com/OSGeo/PROJ/commit/9488b26ca5b7ea763db4e210b311bcf3f04233ca which might have effect on this. It should mostly be a no-op for default builds, although I see that line https://github.com/OSGeo/PROJ/commit/9488b26ca5b7ea763db4e210b311bcf3f04233ca#diff-8a6a421339db369cb2b34097f1a6fede57afb13e7e858e255ed7d6895f814540R1478 might have subtle effects if the proj_lib directory hardcoded at build time doesn't exist at runtime. Previously it will have been used even if not existing, whereas now it is not considered (and thus a proj.db in the current working directory might be used). The new behavior makes more sense to me.

Even

Hi all,

Here is the list of bug fixes in rasterio 1.2.4: 

https://github.com/mapbox/rasterio/blob/master/CHANGES.txt#L13-L32

The wheels on PyPI for this version now include GDAL 3.3.0 and GEOS 3.9.1. The PROJ version remains the same, 7.2.1 until we get more experience with changes to web mercator projection (I believe I see expanded limits) and discovery of the proj.db file.

Thanks for the bug reports and discussions, everyone! 

--

Sean Gillies

-- 
http://www.spatialys.com
My software is free, but my time generally not.