Thanks @ngoldbaum. Should we backport this?

Thanks @ngoldbaum, @mattip. I hope it's not gonna be a long tail of these, as you suspect, but there is still an issue if the r-value is a list:

    a = np.array(['', ''], dtype='T')
    a[[0, 1]] = ['a'*16, 'b'*16]
    print(a)  # getting ['', '']

Note that both a[[0, 1]] = ['a', 'b'] and a[[0, 1]] = np.array(['a'*16, 'b'*16], dtype='T') work (the latter one thanks to the current fix).

And worse, if a has some non-empty values, I get a crash (I am on Windows):

    a = np.array(['a', 'b'], dtype='T')
    a[[0, 1]] = ['a'*16, 'b'*16] # crash

@mattip I added the backport label, thanks for the reminder.

@egesip I opened #27053 to track that. Thanks for the report, If you find any more of these please open issues, it's really helpful to get reports like this.

Will do. Thanks a lot for tackling these issues.