Proposal: ValidateMonoLiterals - Initial bikeshed discussion

[Adam Gundry]

Hi Merijn,

Thanks for persevering with explaining things to me. :-)


On 09/02/15 09:47, Merijn Verstraaten wrote:
>> On 6 Feb 2015, at 21:31, Adam Gundry <adam at well-typed.com> wrote: 
>> What does "all monomorphic cases" mean? Is the idea what GHC would
>> typecheck an expression involving a literal integer, determine that
>> the occurrence had type Even, then evaluate the TH splice *after* 
>> typechecking? Whereas if the occurrence had a non-ground type, it
>> would do something else?
> 
> Yes, Typed TH already runs *after* type checking, which is what
> allows you to do validation based on the resulting type. The main
> reason why I was only proposing to do this for monomorphic values is,
> because, how could you possible validate a polymorphic literal? Which
> validation function would you use?
> 
> You could ban polymorphic literals, but that'd involve eliminating
> most uses of polymorphic Num functions (as I mentioned as another
> email), which would break so much code it would render the extension
> unusable in "real" code. I'm open to better ideas on how to tackle
> this, the main reason I started this discussion is because I don't
> really like this "polymorphic literals fail at compile time" thing
> either. I just don't see how to solve it without going all dependent
> types on the problem.

In the absence of a coherent story for polymorphism, I think the right
thing to do is to be able to specify a particular validator, rather than
try to have type inference determine a monomorphic type and otherwise
get stuck...


>> None of this is particularly persuasive, I'm afraid. Is it
>> worthwhile introducing something new just to avoid having to write
>> a quasi quote?
> 
> Actually, I would be mildly ok with quasi quoters, BUT there
> currently is no Typed TH quasi quoter (as mentioned on the wiki
> page), additionally, such a quoter does not have access to Lift
> instances for all but a handful of datatypes until we have a more
> comprehensive way to generate Lift instances. I think both of these
> points are also highly relevant for this dicussion.

...so is the right solution to introduce Typed TH quasiquoters for
expressions? Sorry, I presumed such a thing existed, as Typed TH is
rather regrettably underdocumented. Is there any particular difficulty
with them, or is it just a Small Matter of Programming?

I think the lack of Lift instances is a separate problem; while it looks
like 7.10 will be better in this respect and dataToExpQ goes a fair way,
I agree that making them easier to generate would be nice. Perhaps a
generics-based default method combined with DeriveAnyClass would make
"deriving Lift" possible?

Adam

-- 
Adam Gundry, Haskell Consultant
Well-Typed LLP, http://www.well-typed.com/