[isabelle-dev] Problem with code generation for non-executable types
Florian Haftmann
florian.haftmann at informatik.tu-muenchen.de
Thu Jan 14 11:47:28 CET 2016
Hi Johannes, hi Andreas,
the core of the problem is that dictionaries are always generated for
type class instances, even if the operations therein are never referred
to. Is this the first time that we have such a situation, or are there
more examples (e.g. in the AFP)? If yes, the code generator could be
equipped to treat particular constants *not* as class parameters.
Cheers,
Florian
Am 11.01.2016 um 12:00 schrieb Johannes Hölzl:
>
>
> Am Samstag, den 09.01.2016, 17:22 +0100 schrieb Florian Haftmann:
>> Ho Johannes.
>>
>>>>> Then the dictionary construction for type constructors
>>>>> does
>>>>> not
>>>>> work in ML! The type signature would be the following:
>>>>>
>>>>> val test_prod : ('a * 'b) filter
>>>>>
>>>>> Which apparently is not allow in ML.
>>>> This is the famous value restriction (which I also regularly run
>>>> into). The standard
>>>> solution is to add a unit closure, because functions may be
>>>> polymorphic in ML.
>>
>> Nothing to add about that.
>>
>> In the examples there is also the problem that constructing a
>> dictionary
>> provokes an exception already. Here the general solution is to hide
>> the
>> problematic terms beneath an abstraction beneath a constructor.
>>
>> Applying that to your examples, I had a look at the sources and came
>> to
>> the conclusion that it is a bad idea have Abs_filter and Rep_filter
>> in
>> generated code at all.
>>
>> For the simple examples, it is much better to use »principal« as a
>> formal constructor, which maps nicely to sets and provides some
>> executability for a considerable class of filters.
>>
>> I did not have an idea in which algebraic framework »uniformity«
>> could
>> fit. Hence I provided a constructor which is a variant of identity
>> and
>> used that, which makes also the examples involving uniformity
>> compileable (but of course not evaluable).
>>
>> Maybe you have an idea how this could be improved.
>
> Well filters are mostly non-computable. Actually I would prefer to tell
> the code-generator to not generate code for topologies and uniform
> spaces at all, as these type classes are carry only non-computable
> data.
>
> But of course your implementation looks cleaner, so I changed in
> Isabelle df65f5c27c15.
>
> - Johannes
>
>
>> Cheers,
>> Florian
>>
>>>
>>> Ah thanks! This explains it.
>>>
>>> Unfortunately, in my case the type is fixed in HOL to:
>>>
>>> uniformity :: ('a * 'a) filter (where 'a :: uniform_space)
>>>
>>> And I do not want to change the type signature for code generation.
>>> So
>>> I will stick to Solution 3.
>>>
>>> - Johannes
>>>
>>>
>>>>> 2. If your type class contains non-computable data, i.e.
>>>>>
>>>>> instantiation bool :: test_class
>>>>> begin
>>>>>
>>>>> definition "test = if P = NP then top else bot"
>>>>>
>>>>> instance proof qed
>>>>> end
>>>>>
>>>>> You get a non-terminating program at the time point when
>>>>> the
>>>>> dictionary for bool :: test_class is constructed. When the
>>>>> code equation is hidden with [code del] one gets a
>>>>> exception!
>>>>>
>>>>> 3. Our current solution is to introduce code_datatype
>>>>> Abs_filter
>>>>> Rep_filter [code del] and define for each uniformity:
>>>>>
>>>>> uniformity = Abs_filter (%P. Code.abort (STR''FAILED'')
>>>>> (Rep_filter test P))
>>>>>
>>>>> @Florian: is the an alternative solution?
>>>>>
>>>>>
>>>>> - Johannes
>>>>>
>>>>> PS: Here is a short, concrete example:
>>>>>
>>>>> theory Scratch
>>>>> imports Complex_Main
>>>>> begin
>>>>>
>>>>> class test_class =
>>>>> fixes test :: "'a filter"
>>>>>
>>>>> instantiation prod :: (test_class, test_class) test_class
>>>>> begin
>>>>>
>>>>> definition [code del]: "test = (test ×⇩F test :: ('a × 'b)
>>>>> filter)"
>>>>>
>>>>> instance proof qed
>>>>> end
>>>>>
>>>>> instantiation unit :: test_class
>>>>> begin
>>>>>
>>>>> definition [code del]:
>>>>> "(test :: unit filter) = bot"
>>>>>
>>>>> instance proof qed
>>>>> end
>>>>>
>>>>> definition "test2 (x::'a::test_class) = (Suc 0)"
>>>>> definition "test3 = test2 ((), ())"
>>>>>
>>>>> value [code] "test3"
>>>>>
>>>>> section ‹Solution›
>>>>>
>>>>> code_datatype Abs_filter
>>>>> declare [[code abort: Rep_filter]]
>>>>>
>>>>> lemma test_Abort: "test = Abs_filter (λP. Code.abort (STR
>>>>> ''test'')
>>>>> (λx. Rep_filter test P))"
>>>>> unfolding Code.abort_def Rep_filter_inverse ..
>>>>>
>>>>> declare test_Abort[where 'a="'a::test_class * 'b ::
>>>>> test_class",
>>>>> code]
>>>>> declare test_Abort[where 'a=unit, code]
>>>>>
>>>>> end
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> Am Freitag, den 08.01.2016, 09:56 +0100 schrieb Johannes Hölzl:
>>>>>> Hi,
>>>>>>
>>>>>> I want to introduce uniform spaces in HOL, for this I need to
>>>>>> introduce
>>>>>> a type class of the form (see the attached bundle):
>>>>>>
>>>>>> class uniformity =
>>>>>> fixes uniformity :: "('a × 'a) filter"
>>>>>>
>>>>>> Note that uniformity is a filter and not a function.
>>>>>>
>>>>>> which sits between topological spaces and metric spaces, i.e.
>>>>>> every
>>>>>> metric type is also in the following type classes:
>>>>>>
>>>>>> class open_uniformity = "open" + uniformity +
>>>>>> assumes open_uniformity: "⋀U. open U ⟷ (∀x∈U. eventually
>>>>>> (λ(x',
>>>>>> y). x' = x ⟶ y ∈ U) uniformity)"
>>>>>>
>>>>>> class uniformity_dist = dist + uniformity +
>>>>>> assumes uniformity_dist: "uniformity = (INF e:{0 <..}.
>>>>>> principal
>>>>>> {(x, y). dist x y < e})"
>>>>>>
>>>>>> Everything works fine until Affinite_Arithmetic, there in
>>>>>> Intersection.thy the code generation fails with the following
>>>>>> ML
>>>>>> error:
>>>>>>
>>>>>> Error: Type mismatch in type constraint.
>>>>>> Value: {uniformity = uniformity_proda} : {uniformity:
>>>>>> 'a}
>>>>>> Constraint: ('a * 'b) uniformity
>>>>>> Reason:
>>>>>> Can't unify 'a to (('a * 'b) * ('a * 'b)) filter
>>>>>> (Type variable is free in surrounding scope)
>>>>>> {uniformity = uniformity_proda} : ('a * 'b) uniformity
>>>>>> At (line 1619 of "generated code")
>>>>>> Exception- Fail "Static Errors" raised
>>>>>>
>>>>>> I assume this is a strange interaction btw code_abort and the
>>>>>> fact
>>>>>> that
>>>>>> uniformity is a filter (datatype 'a filter = _EMPTY) and not
>>>>>> a
>>>>>> function.
>>>>>>
>>>>>> Does anybody know how to avoid such kind of errors? Do I need
>>>>>> to
>>>>>> sprinkle more [code del] or code_abort annotations?
>>>>>>
>>>>>> - Johannes
>>>>>>
>>>>>> _______________________________________________
>>>>>> isabelle-dev mailing list
>>>>>> isabelle-dev at in.tum.de
>>>>>> https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinf
>>>>>> o/is
>>>>>> abel
>>>>>> le-dev
>>>>> _______________________________________________
>>>>> isabelle-dev mailing list
>>>>> isabelle-dev at in.tum.de
>>>>> https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/
>>>>> isab
>>>>> elle-dev
>>>>>
>>> _______________________________________________
>>> isabelle-dev mailing list
>>> isabelle-dev at in.tum.de
>>> https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isab
>>> elle-dev
>>>
>>
> _______________________________________________
> isabelle-dev mailing list
> isabelle-dev at in.tum.de
> https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev
>
--
PGP available:
http://isabelle.in.tum.de/~haftmann/pgp/florian_haftmann_at_informatik_tu_muenchen_de
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: OpenPGP digital signature
URL: <https://mailman46.in.tum.de/pipermail/isabelle-dev/attachments/20160114/f8f7ec74/attachment.sig>
More information about the isabelle-dev
mailing list