Regression in 2.6.4.2 concerning `with`
mechvel opened this issue · comments
This is on agda-2.6.4.2 built by
$ cabal install -foptimise-heavily -j1,
used with stdlib-2.0,
ghc-9.0.2, under Linux Debian 12, x86_64 machine.
The below module fragment is type-checked by agda-2.6.4.1 but not by agda-2.6.4.2 :
private
-- This is the internal loop in compose-reduceMutually
--
aux-I : ℕ → List Pol → List Pol → List Pol
aux-I 0 _ fs = fs
aux-I (suc cnt) fs' fs with any? lmOrdPair? (zip fs' fs)
... | no _ = fs
... | yes _ = aux-I cnt (reduceMutually fs') fs'
compose-reduceMutually : ℕ → List Pol → List Pol
compose-reduceMutually m fs = unzero (aux-I m (reduceMutually fs) fs)
private
-- We do not see how to avoid using these "-base" and "-step" functions.
--
aux-I-base : ∀ m fs' fs → ¬ Any LmOrdPair (zip fs' fs) → aux-I m fs' fs ≡ fs
aux-I-base 0 _ _ _ = PE.refl
aux-I-base (suc m) fs' fs ¬anyOrd with any? lmOrdPair? (zip fs' fs)
... | no _ = PE.refl -- this is the point, line 245 ***
... | yes anyOrd = contradiction anyOrd ¬anyOrd
The command is
$ agda $agdaLibOpt $agdaFlags GBasis/OverEuclidean/I.agda +RTS -M7G
where $agdaFlags = --auto-inline --guardedness.
The report is
home/mechvel/inAgda/doconA/3.2/source/GBasis/OverEuclidean/I.agda:245,51-58
aux-I E vars ppo (suc m) fs' fs
| any? lmOrdPair? (Data.List.zipWith _,_ fs' fs) != fs of type
List
(Pol.OverDecComMonoid.Pol (OfMonoids.mkDecComMonoid
...
(EuclideanDomain.decIntegralDomain E)))))) _≟_)
(Data.List.foldr (λ _ → suc) 0 vars) ppo)
when checking that the expression PE.refl has type
(aux-I E vars ppo (suc m) fs' fs | any? lmOrdPair? (Data.List.zipWith _,_ fs' fs)) ≡ fs
Which Agda is more correct in this case: 2.6.4.1 or 2.6.4.2 ?
Is this fragment sufficient to guess of the source?
It not, may be I would try to simplify the example, (which is a very complex code, using many modules).
It not, may be I would try to simplify the example, (which is a very complex code, using many modules).
Thanks for the report.
Yes, a self-contained example would enable us to start investigating this.
Here is a small self-contained test. It is type-checked by Agda-2.6.4.1 but not by 2.6.4.2.
The details are given in my initial message.
Test.agda.zip
Thanks and congrats, you found a regression in the just-released version of Agda!
A bit unfortunately, 2.6.4.2-rc1 could still handle this example, so a last-minute merge caused this.
In your example, with abstraction of any? lmOrdPair? (zip fs' fs) does not fire any more.
I am afraid changes to normalization in the following PR are the probable cause:
CC: @jespercockx
Here is a shrinking not needing the standard library:
open import Agda.Builtin.Equality
open import Agda.Builtin.Nat
open import Agda.Builtin.List
open import Agda.Builtin.Sigma
data ⊥ : Set where
data Dec (A : Set) : Set where
yes : A → Dec A
no : (A → ⊥) → Dec A
zipWith : ∀{A B C : Set} → (A → B → C) → List A → List B → List C
zipWith f (x ∷ xs) (y ∷ ys) = f x y ∷ zipWith f xs ys
zipWith f _ _ = []
zip : ∀{A B : Set} → List A → List B → List (Σ A λ _ → B)
zip = zipWith (_,_)
Mon = Σ Nat λ _ → Nat
Pol = List Mon
PolPol = Σ Pol λ _ → Pol
postulate
ANY : ∀ {A : Set} → A
_<lm_ : (f g : Pol) → Set
_<lm?_ : ∀ x y → Dec (x <lm y)
Any : ∀{A : Set} → (A → Set) → List A → Set
LmOrdPair : PolPol → Set
LmOrdPair (f , g) = f <lm g
lmOrdPair? : ∀ p → Dec (LmOrdPair p)
lmOrdPair? (f , g) = f <lm? g
any? : ∀{A}{P : A → Set} → (∀ x → Dec (P x)) → ∀ xs → Dec (Any P xs)
any? P? [] = no ANY
any? P? (x ∷ xs)
with P? x | any? P? xs
... | yes p | _ = yes ANY
... | no ¬p | yes q = yes ANY
... | no ¬p | no ¬q = no ANY
{-# TERMINATING #-}
aux-I : List Pol → List Pol → List Pol
aux-I fs' fs
with any? lmOrdPair? (zip fs' fs)
... | no _ = fs
... | yes _ = aux-I fs' fs'
aux-I-base : ∀ fs' fs → aux-I fs' fs ≡ fs
aux-I-base fs' fs
with any? lmOrdPair? (zip fs' fs)
... | yes anyOrd = ANY
... | no _ = reflMore shrinking, getting rid of the builtin libraries:
data _≡_ {A : Set} (x : A) : A → Set where
refl : x ≡ x
record Wrap (A : Set) : Set where
-- pattern; no-eta-equality -- can't turn off eta
constructor wrap
field wrapped : A
data Dec (A : Set) : Set where
no : Dec A
postulate
foo : ∀{A : Set}{P : A → Set} → (∀ x → Dec (P x)) → ∀ (x : A) → Dec (P x)
-- can't inline P = LtWrap
-- can't define foo = id
Nat : Set
P : Nat → Set
p? : ∀ x → Dec (P x)
-- can't make these postulates
PWrap : Wrap Nat → Set
PWrap (wrap f) = P f
pWrap? : ∀ p → Dec (PWrap p)
pWrap? (wrap f) = p? f
bar : Wrap Nat → Wrap Nat
bar fs with foo pWrap? fs
... | no = fs
test : ∀ fs → bar fs ≡ fs
test fs with foo pWrap? fs
... | no = reflProblem seems to be connected to eta for Wrap.