On Tue, Aug 6, 2013 at 11:12 AM, <[email protected]> wrote: > Dear all > > I’ve read on an old status blog > (http://morepypy.blogspot.ch/2012/04/numpy-on-pypy-progress-report.html): > > > We merged record dtypes support. The only missing dtypes left are complex > (important), datetime (less important) and object (which will probably never > be implemented because it makes very little sense and is a mess with moving > GCs). > > > I use object arrays extensively in CPython, because the numpy API is very > convenient and object array operations (such as indexing with an array) can > be much faster than equivalent list comprehensions. I’m measured a > particular indexing operation to be 10x faster than a list comprehension in > CPython. In PyPy, the list comprehension has roughly the same speed as > CPython’s numpy indexing, so there’s nothing to be gained, *except for code > portability*. > > So from a portability point of view, object arrays do make a lot of sense, > even if the implementation may internally rely on lists, because PyPy > optimizes their speed disadvantage away. > > Best regards > Martin > > > Martin Gfeller > ___________________________________________________________________________ > Senior Manager > Head of Quantax > > [email protected] > Xing
Hi Martin We decided we'll implement object arrays, mostly just because why not (and for reasons you mentioned), however you won't be able to take an address of an object array (just like you can in the case of a normal array) and internally use a list. The blog post is simply outdated. Cheers, fijal _______________________________________________ pypy-dev mailing list [email protected] http://mail.python.org/mailman/listinfo/pypy-dev
