cduce issueshttps://gitlab.math.univ-paris-diderot.fr/cduce/cduce/-/issues2022-08-25T15:57:39+02:00https://gitlab.math.univ-paris-diderot.fr/cduce/cduce/-/issues/33Issue with min_type and max_type2022-08-25T15:57:39+02:00Mickael LaurentIssue with min_type and max_typeThe description of the functions Type.Subst.min_type and Type.Subst.max_type in the doc is wrong:
the largest subtype of any instance of t is NOT t where all positive occurrences of variables are replaced by 𝟘 and all negative occurrence...The description of the functions Type.Subst.min_type and Type.Subst.max_type in the doc is wrong:
the largest subtype of any instance of t is NOT t where all positive occurrences of variables are replaced by 𝟘 and all negative occurrences of variables not in vars are replaced by 𝟙.
Counterexample:
`t = (Any->Any) & 'a | (Int->Int) & ~'a`
min(t) should be `(Any->Any)&(Int->Int)`, but replacing the occurences of 'a as specified would give Empty,
which is a subtype of any instance of t but not the largest.
Moreover, the result of this computation is not preserved by semantic equivalence, as for any `'a` we have `(Int->Int) = (Int->Int) & 'a | (Int->Int) & ~'a` but `min(Int->Int) = Int->Int` and `min((Int->Int) & 'a | (Int->Int) & ~'a) = Empty`. Also, I am not sure Cduce would always simplify the type `(Int->Int) & 'a | (Int->Int) & ~'a` into `(Int->Int)` as it does not simplify the DNF of functions.
The cases where the functions min_type and max_type are not correct is when a variable appears in both a negative and positive occurence, because then the operation of substituting one of the occurence by something and the other by something else is not preserved by semantic equivalence.
I understand that one of the usecase of this function is to implement the min and max for dynamic types, but as dynamic type variables should always be unique (a given dynamic type variable should not appear twice in a type), the function clean_type should be enough to implement it (as a given dynamic type variable can only be positive or negative but not both). If a dynamic type variable "semantically" appears in both covariant and contravariant positions, it is invalid and thus the fact that clean_type will do nothing is not an issue (assuming that Cduce will simplify trivial partitions of the form `t & 'a | t & ~'a` in order to only keep occurences of variables that have a semantic impact, which I don't think is always true when t is a function for instance).
My suggestion is to remove the min_type and max_type functions as the cases where it is correct (when no variable is both in covariant and contravariant position) can easily be handled with clean_type.https://gitlab.math.univ-paris-diderot.fr/cduce/cduce/-/issues/26Weak polymorphism is not unified nor propagated when used in a function2022-02-01T17:00:39+01:00MattiasWeak polymorphism is not unified nor propagated when used in a functionSome examples here:
```ocaml
# let r = ref ([ 'a* ]) [];;
val r : { get=[ ] -> [ '_weak0* ] set=[ '_weak0* ] -> [ ] } = { get=<fun>
# let f (x : Int) : Int = r := [ x ]; x + 1;;
val f : Int -> Int = <fun>
# r;;
- : { get=[ ] -> [ '_...Some examples here:
```ocaml
# let r = ref ([ 'a* ]) [];;
val r : { get=[ ] -> [ '_weak0* ] set=[ '_weak0* ] -> [ ] } = { get=<fun>
# let f (x : Int) : Int = r := [ x ]; x + 1;;
val f : Int -> Int = <fun>
# r;;
- : { get=[ ] -> [ '_weak0* ] set=[ '_weak0* ] -> [ ] } = { get=<fun>
set=<fun> }
```
```ocaml
# let g (x : 'a) : 'a = r := [ x ]; x;;
val g : 'a -> 'a = <fun>
# g 3;;
- : 3 = 3
# g;;
- : 'a -> 'a = <fun>
# r;;
- : { get=[ ] -> [ '_weak0* ] set=[ '_weak0* ] -> [ ] } = { get=<fun>
set=<fun> }
```Kim NguyễnKim Nguyễnhttps://gitlab.math.univ-paris-diderot.fr/cduce/cduce/-/issues/25Issue in apply that makes it not monotone2022-01-30T14:35:33+01:00Mickael LaurentIssue in apply that makes it not monotoneWhen intersecting an arrow type whose dnf contain several conjunctions,
if this intersection makes one of the conjunctions empty, it will not be simplified in the DNF.
Then, using the operator apply on this arrow type will consider the ...When intersecting an arrow type whose dnf contain several conjunctions,
if this intersection makes one of the conjunctions empty, it will not be simplified in the DNF.
Then, using the operator apply on this arrow type will consider the positive part of this empty conjunction and thus it will not yield the expected type but a larger one.
This can be reproduced with the following code:
```ocaml
module CD = Cduce_types
let cup t1 t2 = CD.Types.cup t1 t2
let cap t1 t2 = CD.Types.cap t1 t2
let diff = CD.Types.diff
let cons = CD.Types.cons
let mk_arrow = CD.Types.arrow
let true_typ = CD.Builtin_defs.true_type
let false_typ = CD.Builtin_defs.false_type
let bool_typ = cup true_typ false_typ
let int_typ = CD.Types.Int.any
let any = CD.Types.any
let empty = CD.Types.empty
let any_node = cons any
let empty_node = cons empty
let subtype = CD.Types.subtype
let pp_typ = CD.Types.Print.print_noname
let apply t args =
let t = CD.Types.Arrow.get t in
CD.Types.Arrow.apply t args
let arrow_any = CD.Types.Function.any
let domain t =
if subtype t arrow_any then
let t = CD.Types.Arrow.get t in
CD.Types.Arrow.domain t
else empty
let _ =
let simple_arrow = mk_arrow (cons int_typ) (cons int_typ) in
let test = simple_arrow in
let test = cup test (mk_arrow (cons bool_typ) any_node) in
let test = diff test (mk_arrow any_node any_node) in
let test = cap test (mk_arrow (cons (diff any bool_typ)) any_node) in
Format.printf "%a@." pp_typ test ;
Format.printf "Subtype of %a ? %b@." pp_typ simple_arrow (subtype test simple_arrow) ;
Format.printf "But when we apply an Int to it: %a@." pp_typ (apply test (domain simple_arrow)) ;
```
It has been tested on the branches dune-switch and polymorphic.Kim NguyễnKim Nguyễnhttps://gitlab.math.univ-paris-diderot.fr/cduce/cduce/-/issues/17Unsoundness with record typing2021-05-02T10:41:43+02:00Tommaso PetruccianiUnsoundness with record typingFunctions on records can be given unsound types as in the following:
# let f = fun ({..} -> 'a) x -> x;;
val f : Record -> 'a = <fun>
# f {};;
- : 'a = { }
It seems that this `'a` cannot be instantiated:
# (f {}) + 1;;
Characte...Functions on records can be given unsound types as in the following:
# let f = fun ({..} -> 'a) x -> x;;
val f : Record -> 'a = <fun>
# f {};;
- : 'a = { }
It seems that this `'a` cannot be instantiated:
# (f {}) + 1;;
Characters 1-5:
This expression should have type:
!float | { .. } | Int
but its inferred type is:
'a
which is not a subtype, as shown by the sample:
Atom & 'a
However, here it becomes unsound:
# f (f {});;
- : Empty = { }
Typing of field removal is also unsound:
# let g = fun ({a=Int;b=Bool;..}&'a -> {b=Bool;..}&'a) x -> x\a;;
val g : { a=Int b=Bool .. } & 'a -> { b=Bool .. } & 'a = <fun>
# g {a=2;b=`true};;
- : { a=2 b=`true } = { b=true }
# (g {a=2;b=`true}).a;;
- : 2 = true
# (g {a=2;b=`true}).b;;
- : `true = true
# (g {a=2;b=`true}).a + 2;;
File "runtime/value.ml", line 873, characters 9-15: Assertion failedKim NguyễnKim Nguyễnhttps://gitlab.math.univ-paris-diderot.fr/cduce/cduce/-/issues/12Solving type equation with recursive types for a variable 'a does not work wh...2017-10-16T08:48:07+02:00Kim NguyễnSolving type equation with recursive types for a variable 'a does not work when 'a occurs below a union/intersection/negationThe problem is in decompose and solve_rectype in Types. The Positive module giving the least fix point solution of a system of equation (aka Courcelle) does not work when variable occurs below type connective.
See for instance:
```
...The problem is in decompose and solve_rectype in Types. The Positive module giving the least fix point solution of a system of equation (aka Courcelle) does not work when variable occurs below type connective.
See for instance:
```
debug tallying [] [ (Int, 'a),'b ; 'b, 'a ] ;;
```
Kim NguyễnKim Nguyễn