refactor(api): make input value coercion of mutate() identical to select()

[kszucs]

String and integer literals passed to select() are interpreted as columns whereas mutate() interpreted them as literals.

BREAKING CHANGE: strings passed to table.mutate() are now interpreted as column references instead of literals, use ibis.literal(string) to pass the string as a literal

[kszucs]

I am actually not a big fan of t.select(42) trying to locate a column. Treating strings as column references is fine but integers are rather weird.

[cpcloud]

Makes sense. Still a good chunk of failures though.

[jcrist]

I am actually not a big fan of t.select(42) trying to locate a column. Treating strings as column references is fine but integers are rather weird.

I'd agree with this and would propose we deprecate (or just remove) ints as column references everywhere except when directly indexing a table (e.g. t[0]).

I do think that treating everything passed to mutate as a literal is a bit more useful behavior (t.mutate(new_col="existing_col") would be less common IMO than t.mutate(new_col="some_constant")), but I can see the argument here for consistency across methods.

[cpcloud]

+1 to removing the int-as-column references in mutate and select (I can't remember which one has it and which one doesn't), and keeping the string as column references but only in select.

We have rename for the use cases this covers.

[kszucs]

Now both are consistent. My preference is to remove both, but I am afraid that this would be too big of a breaking change. Though we actually break it now, and provide a config option to fall back to the previous behavior.

[jcrist]

+1 to removing the string-and-int-as-column references in mutate and select (I can't remember which one has it and which one doesn't)

IMO we'll still want to support strings as column references. t.select("x", "y", "z") is such a common pattern, and feels intuitive to me (while t.select(0, 1, 2) doesn't).

[kszucs]

Ok, I think we have an agreement on removing the int case. Going to put up a follow-up for that.

[cpcloud]

We need to document this as part of this PR or the next one, because it seems pretty non trivial to articulate what the rules are especially around strings.

What exactly is the behavior of each of str and int without calling literal in each of select and mutate?

[kszucs]

Created a follow-up issue for the documentation #8879