[m-rev.] for review: ensure every option is initialized

[Julien Fischer]

Hi Zoltan,

On Fri, 18 Sep 2020, Zoltan Somogyi wrote:

> The diff itself is boring; I am seeking feedback on the concept.
>
> I am in two minds about this change. On the one hand, it does ensure that
> every option is initialized exactly once, which the old code did not,
> as evidenced by the fact that it caught an uninitialized option. On the other,
> it adds another several hundred lines to an already-too-large module,
> and the new version is not as readable as the old, due to the presence
> of the option category next to every option. We can ensure unique
> initialization only with a switch on the option at the top level, while
> I can't see a way to avoid listing the category next to each option
> without making the top level switch a switch on the category.
> The only resolution I can see is to make the option type
> not a single enum type like this:
>
> :- type option ---> opt1; opt2; ...
>
> but a structured type like this:
>
> :- type option ---> category1(opt_in_cat1) ; category2(opt_in_cat2) ...
>
> In this case, the top level switch on the category is also a switch
> on a (part of) the option itself. However, I am far from sure whether
> such a change is desirable.

My preference is for the first approach.
(The diff looks fine.)

Julien.