You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Theres a milestone referencing this, along with various comments in the scipy source.
Was this a major revision that was decided to be a bad idea and cancelled, or simply a quirk in the version numbering? If the latter, what version did 2.0.0 end up actually being?
Yeah, though my guess is, the "NumPy 2.0" idea has not really been anything more then a possible idea since at least numpy 1.7. So, basically the whole thing is just an archaeological detail at this time.
There was discussion of an API breaking refactor of NumPy for some time back in the early days, and there have been three attempts at what might be labeled NumPy 2.0 that I know of. The first was Enthought, funded by Microsoft, back around 2010-2011, the numpy-refactor repository here is a fork of the Enthought work. @mwiebe had a plan to move NumPy forward, involving moving to C++, expanding dtypes, and much else, and that foundered after the missing values kerfuffle. That kerfuffle pretty much put an end to NumPy making major compatibility breaks. Mark continued developing his ideas on the side during his stint at Continuum, Blaze is the result.
In case you are curious, the missing values idea was essentially to move masking into the ndarray base class. The ufunc where keyword is one of several artifacts of that.
That is the short history as I recall it. @teoliphant could probably fill in some details and pin down the dates. At this point, NumPy 2.0 is a creature of mythology.
Activity
pv commentedon May 6, 2017
I think "Numpy 2.0" just informally means a hypothetical, backward-compatibility breaking redesign of Numpy somewhere in the far future.
seberg commentedon May 6, 2017
Yeah, though my guess is, the "NumPy 2.0" idea has not really been anything more then a possible idea since at least numpy 1.7. So, basically the whole thing is just an archaeological detail at this time.
charris commentedon May 6, 2017
There was discussion of an API breaking refactor of NumPy for some time back in the early days, and there have been three attempts at what might be labeled NumPy 2.0 that I know of. The first was Enthought, funded by Microsoft, back around 2010-2011, the numpy-refactor repository here is a fork of the Enthought work. @mwiebe had a plan to move NumPy forward, involving moving to C++, expanding dtypes, and much else, and that foundered after the missing values kerfuffle. That kerfuffle pretty much put an end to NumPy making major compatibility breaks. Mark continued developing his ideas on the side during his stint at Continuum, Blaze is the result.
In case you are curious, the missing values idea was essentially to move masking into the ndarray base class. The ufunc
wherekeyword is one of several artifacts of that.That is the short history as I recall it. @teoliphant could probably fill in some details and pin down the dates. At this point, NumPy 2.0 is a creature of mythology.
shoyer commentedon May 6, 2017
While at one point this was called "Blaze", Mark's work lives on in as DyND. The Python Blaze project no longer uses DyND.
eric-wieser commentedon Oct 13, 2018
Some more info is here: https://web.archive.org/web/20130402032018/http://projects.scipy.org/numpy/roadmap
WarrenWeckesser commentedon Oct 26, 2021
I think the question raised in this issue has been answered, so closing.