Merge pull request #367 from mpiannucci/netcdf3-padding

Handle padding in netcdf3 translation

kerchunk/netCDF3.py

@@ -207,9 +207,15 @@ def translate(self):
             outer_shape = size // dt.itemsize
             offset = start
             for name in dt.names:
+                dtype = dt[name]
+
+                # Skip padding, but increment offset.
+                if name.startswith("_padding_"):
+                    offset += dtype.itemsize
+                    continue
+
                 # the order of the names if fixed and important!
                 var = self.variables[name]
-                dtype = dt[name]
                 base = dtype.base  # actual dtype
                 shape = (outer_shape,) + dtype.shape
 

kerchunk/tests/test_netcdf.py

@@ -45,6 +45,11 @@ def unlimited_dataset(tmpdir):
     rootgrp.createDimension("lat", 10)
     rootgrp.createDimension("lon", 5)
     rootgrp.createVariable("time", "f8", ("time",))
+    # reference time is an unbounded dimension that is a half byte long, so it
+    # has padding to line up to take up exactly one byte. It is here to test that
+    # kerchunk can handle the padding correctly and read following variables
+    # correctly.
+    rootgrp.createVariable("reference_time", "h", ("time",))
     rootgrp.title = "testing"
     latitudes = rootgrp.createVariable("lat", "f4", ("lat",))
     longitudes = rootgrp.createVariable("lon", "f4", ("lon",))