full works on lists but with very bad performance

When you pass a python list to full, it actually works as expected but with bad performance. The documentation says that only a scalar is accepted, but because it does the right thing, it’s very easy for people to end up with code like this that’s correct but inefficient.

import numpy
x = numpy.full(100, range(100))

The reason it’s inefficient is because the source looks like this:

    if dtype is None:
        dtype = array(fill_value).dtype
    a = empty(shape, dtype, order)
    multiarray.copyto(a, fill_value, casting='unsafe')
    return a

If the dtype is not specified, the branch actually creates a copy of the whole array, throws it away, and then does it again.

I would suggest either explicitly supporting this use case, and improving performance and docs to match. Or, try to do some kind of basic check to reject this input as unacceptable.

[ghost]

Thanks all for the help, @eric-wieser I just created a new branch and did the changes again since it was such a small change.

[ghost]

Ah thank you! Just pulled the trigger on that. Apologies for the hold up, I’m new to this workflow and thought I was done after simply doing a commit.