Closed lnicola closed 9 months ago
Will rebase on top of #489.
r? @metasim
Should we also support {read|write}_block
without ndarray
?
Should we also support {read|write}_block without ndarray?
I think we should, but I don't think I want to do it in this PR, since there's a bit of design space to explore. I don't like the current situation either, but in the meanwhile it's probably better than having no access to these.
@metasim do you ever use the MD API, and would you expect it to require ndarray
, or use some form of Buffer3
, BufferN
version?
do you ever use the MD API, and would you expect it to require ndarray, or use some form of Buffer3, BufferN version?
@lnicola I do not. That's a good point (I didn't realize MD required it). For that API it seems reasonable to require ndarray
, and would be silly to implement a bespoke multidimensional array capability just to avoid a dependency.
I wouldn't be adverse to ndarray
being required for everything. But because we advertise it as an "optional" dependency, I don't think not using it should preclude access to basic functionality (i.e. block I/O).
I didn't realize MD required it
It doesn't, I should have checked before asking :pensive:.
PS: I think ndarray
is fantastic. My position is more about it's optionality, and treating Buffer
as a first-class citizen in face of that optionality.
[x] I added an entry to
CHANGES.md
if knowledge of this change could be valuable to users.