dask / dask-image / 127 / 2

100%

main: 99%

push

travis-ci

Handle structured dtype in labeled_comprehension (#66)

* Type cast record values in `labeled_comprehension`

Make sure that type casting within `labeled_comprehension` and
associated utility functions handles record (a.k.a. structured) values
in addition to typical scalar types.

* Scalar arrays in _labeled_comprehension_delayed

Instead of squeezing NumPy arrays into scalars when returning results
from `_labeled_comprehension_delayed`, actually just return NumPy 0-D
scalar arrays. These behave a little better in some cases (e.g. when a
function returns a scalar array).

* Use 1-D singleton array in `labeled_comprehension`

To better handle structured type values that might be returned from the
custom user function, coerce all results returned by
`labeled_comprehension` into 1-D singleton arrays. As Dask Arrays seem
to do a better job of handling structured types when they have some
shape to them, this addresses returning structured types.

* Adjust `labeled_comprehension` result handling

Preallocate an array to store the result in
`_labeled_comprehension_delayed`. Makes sure that writing into the
`default` value doesn't happen by accident. Also makes sure that only
scalar values are handled from custom user functions. As not enforcing
these constraints may result in the computation breaking later in more
confusing ways, explicit handling of them in
`_labeled_comprehension_delayed` should point out to the user where they
may have gone wrong.

* Test `labeled_comprehension` with structured types

Include a test using `labeled_comprehension` with structured dtypes.

209 of 209 branches covered (100.0%)

Branch coverage included in aggregate %.

702 of 702 relevant lines covered (100.0%)

1.0 hits per line