On 18 October 2016 at 14:24, Ashish Agarwal <agarwal1...@gmail.com> wrote:
> On Tue, Oct 18, 2016 at 4:38 AM, Daniel Bünzli <daniel.buen...@erratique.ch>
> wrote:
>> I think you should transform your errors according to your level of
>> abstraction as much as you do with your values.
> That's a good goal.

Yes; I think it fits very well with the point of view I think
previously expressed by one (or both :) of you that the distinction
Ok/Error is rather false-- the problem is actually to be able to treat
subsets of possible return values in aggregate, rather than the tags
`Ok or `Error having any special meaning.

>> Also using inclusion (by tagging an error subset) rather than union for
>> errors reported by subsystems allows you to work with a constant set of
>> variants at a given abstraction level which lessens the problems.
> I've tried that, and only problem was I ended up with rather big types.
> Appropriate naming could get around that, at least in the mli's. Crazy types
> might still get printed out by utop, but that doesn't bother me too much.

Sorry-- not familiar enough with the terminology. Any chance one of
you could point to an example of using "inclusion rather than union"


Richard Mortier

MirageOS-devel mailing list

Reply via email to