Max is broken about awkward-0.x HOT 1 CLOSED

Take a look at the diff and have a laugh.

Actually, for something.max().max(), something.flatten().max() would always be equal and less effort to compute.

I should probably start publicizing that something.content is not to be used unless you're interested in the internals of your array. (It doesn't start with an underscore; there seems to be a third level between "public" and "private" for starts/stops/content.) The content need not be contiguous and it can contain unreachable elements. Two arrays might look the same and result in the same list of lists with .tolist() (i.e. the same "logical structure") but one has elements in its content that the other doesn't have (i.e. different "physical structure"). For instance, do

something[:, 1:]

to logically drop the first element in each sublist and you'll get a new starts but the same content. The elements that have been dropped from the logical view would contribute to content.max().

I'm beginning to realize that this is a gotcha on the order of Numpy's view vs copy.

from awkward-0.x.