[Haskell-cafe] ANNOUNCE: partial-lens 0.0.1

[Erik Hesselink]

On Wed, Dec 21, 2011 at 16:39,  <roconnor at theorem.ca> wrote:
> On Wed, 21 Dec 2011, Erik Hesselink wrote:
>
>> How does this relate to the Maybe lenses in fclabels [1]?
>>
>> Erik
>>
>> [1]
>> http://hackage.haskell.org/packages/archive/fclabels/1.0.4/doc/html/Data-Label-Maybe.html
>
>
> It appears to be somewhere between similar and the same.
>
> *** Comparison of API
>
> Data.Label.Maybe.get corresponds to Data.Lens.Partial.getPL
>
> Data.Label.Maybe.set roughly corresponds to Data.Lens.Partial.trySetPL
> except that trySetPL will bail out early if the reference is null.  We can
> match the signature of set more precisely by:
>
> Data.Label.Maybe.set l v r ~ Data.Lens.Partial.trySetPL l r <*> pure v
>
> Data.Label.Maybe.modify would correspond to Data.Lens.Partial.tryModPL if I
> had implemented it ... which maybe I ought to.
>
> Data.Label.Maybe.embed corresponds to a composition of totalLens and
> maybeLens.  More specifically
>
> Data.Label.Maybe.embed l ~ Data.Lens.Partial.maybeLens .
> Data.Lens.Partial.totalLens l
>
> Data.Label.MaybeM.gets roughly corresponds to
> Data.Lens.Partial.Lazy.accessPlus except that accessPlus is particular to
> StateT because partial-lens is a Haskell 98 compliant package.  I need to
> write partial-lens-fd which will contain a function precisely corresponding
> to Data.Label.MaybeM.gets
>
> I don't have Data.Label.MaybeM.asks, because there was no corresponding
> functionality in data-lens.  We should probably add a version of this.
>
> *** Comparison of representation
>
> The usual differences between data-lens and fclabels applies to partial-lens
> as well.  The representation for data-lens and partial-lens allows modify to
> be done with one case analysis on a record since the getter and setters are
> combined into one coalgebra whereas in fclabels two case analysis must be
> done: one for the getter and one for the setter.  When chains of lenses are
> composed, I'm told the differences become more apparent.
>
> In partial-lens, the combination of getter and setter into a single
> coalgebraic operations means that the getter and setter are statically
> forced to return Nothing on the same record; but this is not enforced with
> the fclabels representation.
>
> That said, perhaps the MaybeLens from fclabels is trying to do something
> different.  I don't know what laws you expect to hold for the getter and
> setters of a maybe lens since it isn't documented (actually I appear to have
> also forgotten to document the coalgebra laws for a comonad in my package)
> so perhaps MaybeLens are intended to be more general than partial lenses.
>
> For example maybe a user wants to make it illegal to set the birth date to
> be greater than the death date in a record.  In this case getting the birth
> date will succeed, but setting will fail if the provided birth date out of
> bounds.  This is possible to write using MaybeLens, but is impossible with
> partial lenses since with partial-lenses either the reference is null,
> meaning getting and setting both fail, or it is not null which means that
> getting and setting both succeed.

Thanks for the detailed explanation! It seems they are indeed (almost)
the same, apart from the differences in representation.

The original motivation for the Maybe lenses in fclabels was accessing
record fields with Maybe types and composing these lenses (even in the
presence of multiple Maybes). It does not come from a category
theoretical starting point, hence no laws (yet). Your final example is
interesting, I'd never considered doing something like that.

Erik