readbuf size, despite profiling showing that it doesn't make much
difference.

of lenght 1.

keyword arguments.

TaskThread also doesn't crash when an enqueued task raises an error.

a new thread.

2. ImageWrapper marks its expand coverage task thread as a daemon.

because __taskThread attr does not exist.

optionally be told to calculate data ranges on a separate thread.

to collapse many little expansions (that cover individual volumes) into
a few larger ones (that cover many volumes).

equivalent expansions which pertain to adjacent vectors/slices/volumes.

Bugfix in ImageWrapper.calcExpansion - could not handle slice outside of
coverage, but overlapping with some dimensions.

Initial support in Image class for data editing.

regions are specified with nan (which can't be represented as uint32).

documentation.

into a separate method, as it will probably come in handy for write
support.

offers a more fine-grained result. Adjusted sliceCovered - function it
doesn't need to know the image shape. updateDataRangeOnRead is no longer
memoized - slice coverage is checked before it is called, so it
shouldn't need to be memoized.

also skips __getitem__ coverage update if the full image data range has
been calculated.

between expansions.

no sense, I know.

data is read. But the dataRange property will return cal_min/cal_max
if no data has been read. Some documentation added too.

comparisons to None now testing for isnan.

ImageWrapper now maintains separate coverages for each volume (in a 4D
image, or slice in a 3D image), and adjusts them individually. Lots of
crazy code which seems to work ok, but has not been tested or
documented. Most of the complex stuff has been turned into standalone
functions, so can be unit-tested.

changes. Bugfix in Image.__getitem__ - was potentially returning data
with trailing dimensions of size 1.