Support a start_index of 1

[Huite]

No description provided.

[veenstrajelmer]

This works like a charm, writing a netfile with start_index=1 is now much simpler than before. I have tested:

• writing a file from scratch (start_index can conveniently be set before writing)
• reading an existing file and writing it again (properties are maintained)
• reading an existing file, cropping the grid and writing it again (properties are maintained)

Thanks a lot for this useful improvement! The diff also shows quite some fill_value changes, I guess these are additions to #299 right? I did not specifically look into the behaviour of fill_value, but just let me know if I should.

[Huite]

The diff also shows quite some fill_value changes, I guess these are additions to #299 right? I did not specifically look into the behaviour of fill_value, but just let me know if I should.

Some fill value logic is required to avoid increment or decrementing the fill values in the connectivity arrays.

This is one of the few cases where having masked arrays instead of ordinary numpy arrays have some merit:

import numpy as np
import numpy.ma as ma

x = np.array([1, 2, 3, -1, 5])
mx = ma.masked_array(x, mask=[0, 0, 0, 1, 0])
mx2 = mx + 1

print(mx.data)  # array([ 1,  2,  3, -1,  5])
print(mx2.data)  # array([ 2  3  4 -1  6])

When indexing, it'll use the original value:

a = np.arange(10)
assert np.array_equal(a[x], a[mx])

The -1 value is therefore still the best choice, since it will be used for indexing.

It would remove the need to generate is_fill = connectivity == FILL_VALUE.
A nice feature is that masked does propagate during indexing.

Anyway, just a side note.