Thanks! It's only a default anyway, so the question is, which is less surprising? Given that vectors are (rightly, in my view) mutable in Scheme, I think that making arrays mutable is less surprising. Opinions may legitimately differ.
I think that's because you do array operations in bulk using pipelines,
whereas most casual array users are used to mutable arrays
While primitives internally usually have subscripting operations, to end users
of the primities, the bulk pipeline approach, which was popularized by APL,
then fell dormant for a few decades, and then was ressurected starting with
Matlab, then Numpy, and more recently Torch, is natural. Today, at least in
the ML and deep learning communities, a purely functional bulk pipeline
approach reminiscent of APL is the dominant programming style.
Jeff (http://engineering.purdue.edu/~qobi)