Naming of rows() and cols() functions

[yongqli]

This might seem like a nitpick, but I strongly prefer the name nrows() over rows(). I think of rows as referring to the rows objects themselves rather than the count, which is a property they possess. (num_rows() would work for me as well.) Is there any hope of adding nrows() to BaseMatrix as an alias to rows()?

[James Lucas]

You're not the first person to suggest this which means we should give it some serious thought!

If I'm understanding correctly, your suggestion is to:

• Change current rows() to nrows().
• Change current cols() to ncols().
• Change current row_iter() to rows().
• Change current col_iter() to cols().

I think @tafia suggested the same a while ago.

How do others feel about this?

[Andreas Borgen Longva]

I think this is down to different conventions. For example, Eigen uses rows()/cols(), while Armadillo uses n_rows().

I think in any case, it's better to choose a convention and stick to it than to start adding alias functions, which may give precedent for adding other aliases, and suddenly the various traits are bloated with way too many functions.

Personally I prefer rows as it's what I'm used to from Eigen, but from a somewhat more objective angle, nrows doesn't work very well with autocomplete. That is, I think it should start with the word row (i.e. row_count, rows or similar) so that you can easily navigate the API when you're not familiar with it.

I do however oppose renaming row_iter to rows, as I think it's not clear at all that rows is an iterator, and I think being explicit about this is a good idea.

[James Lucas]

Personally I'm with @Andlon . I prefer the current naming (though don't have an issue with having rows() return an iterator).

I am also against adding alias functions, for the reasons @Andlon gave. We should rename the functions if there is an overwhelming consensus though.

[Andreas Borgen Longva]

I think removing ambiguity in names is a good idea in any case. Actually, the more I think about it, the more I like row_count(). It's more verbose, of course, but its purpose is clear as crystal and it's autocomplete-friendly.

I also wanted to make clear that although I prefer row_iter over rows for returning an iterator over the rows, I'm not very strongly opposed to this change, so if there is concensus on that I don't mind. There's already diag which returns an iterator, so it wouldn't be unprecedented.