search

rasterio / rasterio-wheels

MIT License
11 stars 16 forks source link

Try GDAL 3.2.0 and PROJ 7.2.0 #54

Closed sgillies closed 3 years ago

sgillies commented 4 years ago

Bounds densification test is failing: https://travis-ci.com/github/rasterio/rasterio-wheels/jobs/384267867#L2747-L2762.

To date we haven't been able to test https://github.com/mapbox/rasterio + GDAL 3.1 because the GDAL builds time out https://travis-ci.org/github/mapbox/rasterio/jobs/723190297 so I can't say yet if this is a real change in results or a rasterio bug.

drnextgis commented 4 years ago

It doesn't seem a rasterio bug.

GDAL (master branch):

$ gdaltransform -s_srs EPSG:4326 -t_srs EPSG:2163
Enter X Y [Z [T]] values separated by space, and press Return.
-120 40
-1688722.36656949 -350040.508198484 0

GDAL 2.2.3:

$ gdaltransform -s_srs EPSG:4326 -t_srs EPSG:2163
-120 40
-1684649.41338287 -350356.813765847 0
drnextgis commented 4 years ago

But at the same time if define SRS using RPOJ string it gives the same result as for GDAL2:

$ gdaltransform -s_srs EPSG:4326 -t_srs "+proj=laea +lat_0=45 +lon_0=-100 +x_0=0 +y_0=0 +a=6370997 +b=6370997 +units=m +no_defs"
Enter X Y [Z [T]] values separated by space, and press Return.
-120 40
-1684649.41338287 -350356.813765848 0
$ gdalinfo --version
GDAL 3.2.0dev-8f5bab77ee, released 2020/09/13
drnextgis commented 4 years ago

Different versions of GDAL return different PROJ strings for the same EPSG code:

GDAL (master branch):

$ gdalsrsinfo -o proj4 epsg:2163

+proj=laea +lat_0=45 +lon_0=-100 +x_0=0 +y_0=0 +datum=NAD27 +units=m +no_defs

GDAL 2.2.3:

$ gdalsrsinfo -o proj4 epsg:2163
'+proj=laea +lat_0=45 +lon_0=-100 +x_0=0 +y_0=0 +a=6370997 +b=6370997 +units=m +no_defs '
drnextgis commented 4 years ago

EPSG:2163 no longer supported by EPSG because datum information is required for unambiguous spatial referencing. Source: https://epsg.org/crs_2163/US-National-Atlas-Equal-Area.html

drnextgis commented 4 years ago 
# git diff
diff --git a/tests/test_warp.py b/tests/test_warp.py
index df6c26b1..3b031846 100644
--- a/tests/test_warp.py
+++ b/tests/test_warp.py
@@ -221,7 +221,7 @@ def test_transform_bounds_densify():
     # This transform is non-linear along the edges, so densification produces
     # a different result than otherwise
     src_crs = CRS.from_epsg(4326)
-    dst_crs = CRS.from_epsg(2163)
+    dst_crs = CRS.from_proj4("+proj=laea +lat_0=45 +lon_0=-100 +x_0=0 +y_0=0 +a=6370997 +b=6370997 +units=m +no_defs")
     assert np.allclose(
         transform_bounds(src_crs, dst_crs, -120, 40, -80, 64, densify_pts=0),
         (-1684649.41338, -350356.81377, 1684649.41338, 2234551.18559),
$ pytest -q -k 'test_transform_bounds_densify and not test_transform_bounds_densify_out_of_bounds'
Test data present.
.                                                                                                                                                      [100%]
1 passed, 1809 deselected in 4.39s
sgillies commented 3 years ago

Ooh, so very close in https://travis-ci.com/github/rasterio/rasterio-wheels/builds/208241915. Linux wheels look good https://travis-ci.com/github/rasterio/rasterio-wheels/jobs/458767026, but there's a duplicate library error from delocate in the macos case: https://travis-ci.com/github/rasterio/rasterio-wheels/jobs/458767028. Perhaps it's related to https://github.com/matthew-brett/delocate/issues/53. Will dig a little more this weekend.