Theory DetMonad
theory DetMonad
imports "../Lib"
begin
text ‹
\label{c:monads}
State monads are used extensively in the seL4 specification. They are
defined below.
›
section "The Monad"
text ‹
The basic type of the deterministic state monad with failure is
very similar to the normal state monad. Instead of a pair consisting
of result and new state, we return a pair coupled with
a failure flag. The flag is @{const True} if the computation have failed.
Conversely, if the flag is @{const False}, the computation resulting in
the returned result have succeeded.›
type_synonym ('s,'a) det_monad = "'s ⇒ ('a × 's) × bool"
text ‹
The definition of fundamental monad functions ‹return› and
‹bind›. The monad function ‹return x› does not change
the state, does not fail, and returns ‹x›.
›
definition
return :: "'a ⇒ ('s,'a) det_monad" where
"return a ≡ λs. ((a,s),False)"
text ‹
The monad function ‹bind f g›, also written ‹f >>= g›,
is the execution of @{term f} followed by the execution of ‹g›.
The function ‹g› takes the result value \emph{and} the result
state of ‹f› as parameter. The definition says that the result of
the combined operation is the result which is created
by ‹g› applied to the result of ‹f›. The combined
operation may have failed, if ‹f› may have failed or ‹g› may
have failed on the result of ‹f›.
›
text ‹
David Sanan and Zhe Hou: The original definition of bind is very inefficient
when converted to executable code. Here we change it to a more efficient
version for execution. The idea remains the same.
›
definition "h1 f s = f s"
definition "h2 g fs = (let (a,b) = fst (fs) in g a b)"
definition bind:: "('s, 'a) det_monad ⇒ ('a ⇒ ('s, 'b) det_monad) ⇒
('s, 'b) det_monad" (infixl ">>=" 60)
where
"bind f g ≡ λs. (
let fs = h1 f s;
v = h2 g fs
in
(fst v, (snd v ∨ snd fs)))"
text ‹
Sometimes it is convenient to write ‹bind› in reverse order.
›
abbreviation(input)
bind_rev :: "('c ⇒ ('a, 'b) det_monad) ⇒ ('a, 'c) det_monad ⇒
('a, 'b) det_monad" (infixl "=<<" 60) where
"g =<< f ≡ f >>= g"
text ‹
The basic accessor functions of the state monad. ‹get› returns
the current state as result, does not fail, and does not change the state.
‹put s› returns nothing (@{typ unit}), changes the current state
to ‹s› and does not fail.
›
definition
get :: "('s,'s) det_monad" where
"get ≡ λs. ((s,s), False)"
definition
put :: "'s ⇒ ('s, unit) det_monad" where
"put s ≡ λ_. (((),s), False)"
subsection "Failure"
text ‹The monad function that always fails. Returns the current
state and sets the failure flag.›
definition
fail :: "'a ⇒ ('s, 'a) det_monad" where
"fail a ≡ λs. ((a,s), True)"
text ‹Assertions: fail if the property ‹P› is not true›
definition
assert :: "bool ⇒ ('a, unit) det_monad" where
"assert P ≡ if P then return () else fail ()"
text ‹An assertion that also can introspect the current state.›
definition
state_assert :: "('s ⇒ bool) ⇒ ('s, unit) det_monad"
where
"state_assert P ≡ get >>= (λs. assert (P s))"
subsection "Generic functions on top of the state monad"
text ‹Apply a function to the current state and return the result
without changing the state.›
definition
gets :: "('s ⇒ 'a) ⇒ ('s, 'a) det_monad" where
"gets f ≡ get >>= (λs. return (f s))"
text ‹Modify the current state using the function passed in.›
definition
modify :: "('s ⇒ 's) ⇒ ('s, unit) det_monad" where
"modify f ≡ get >>= (λs. put (f s))"
lemma simpler_gets_def: "gets f = (λs. ((f s, s), False))"
apply (simp add: gets_def return_def bind_def h1_def h2_def get_def)
done
lemma simpler_modify_def:
"modify f = (λs. (((), f s), False))"
by (simp add: modify_def bind_def h1_def h2_def get_def put_def)
text ‹Execute the given monad when the condition is true,
return ‹()› otherwise.›
definition
when1 :: "bool ⇒ ('s, unit) det_monad ⇒
('s, unit) det_monad" where
"when1 P m ≡ if P then m else return ()"
text ‹Execute the given monad unless the condition is true,
return ‹()› otherwise.›
definition
unless :: "bool ⇒ ('s, unit) det_monad ⇒
('s, unit) det_monad" where
"unless P m ≡ when1 (¬P) m"
text ‹
Perform a test on the current state, performing the left monad if
the result is true or the right monad if the result is false.
›
definition
condition :: "('s ⇒ bool) ⇒ ('s, 'r) det_monad ⇒ ('s, 'r) det_monad ⇒ ('s, 'r) det_monad"
where
"condition P L R ≡ λs. if (P s) then (L s) else (R s)"
notation (output)
condition ("(condition (_)// (_)// (_))" [1000,1000,1000] 1000)
subsection ‹The Monad Laws›
text ‹Each monad satisfies at least the following three laws.›
text ‹@{term return} is absorbed at the left of a @{term bind},
applying the return value directly:›
lemma return_bind [simp]: "(return x >>= f) = f x"
by (simp add: return_def bind_def h1_def h2_def)
text ‹@{term return} is absorbed on the right of a @{term bind}›
lemma bind_return [simp]: "(m >>= return) = m"
apply (rule ext)
apply (simp add: bind_def h1_def h2_def return_def split_def)
done
text ‹@{term bind} is associative›
lemma bind_assoc:
fixes m :: "('a,'b) det_monad"
fixes f :: "'b ⇒ ('a,'c) det_monad"
fixes g :: "'c ⇒ ('a,'d) det_monad"
shows "(m >>= f) >>= g = m >>= (λx. f x >>= g)"
apply (unfold bind_def h1_def h2_def Let_def split_def)
apply (rule ext)
apply clarsimp
done
section ‹Adding Exceptions›
text ‹
The type @{typ "('s,'a) det_monad"} gives us determinism and
failure. We now extend this monad with exceptional return values
that abort normal execution, but can be handled explicitly.
We use the sum type to indicate exceptions.
In @{typ "('s, 'e + 'a) det_monad"}, @{typ "'s"} is the state,
@{typ 'e} is an exception, and @{typ 'a} is a normal return value.
This new type itself forms a monad again. Since type classes in
Isabelle are not powerful enough to express the class of monads,
we provide new names for the @{term return} and @{term bind} functions
in this monad. We call them ‹returnOk› (for normal return values)
and ‹bindE› (for composition). We also define ‹throwError›
to return an exceptional value.
›
definition
returnOk :: "'a ⇒ ('s, 'e + 'a) det_monad" where
"returnOk ≡ return o Inr"
definition
throwError :: "'e ⇒ ('s, 'e + 'a) det_monad" where
"throwError ≡ return o Inl"
text ‹
Lifting a function over the exception type: if the input is an
exception, return that exception; otherwise continue execution.
›
definition
lift :: "('a ⇒ ('s, 'e + 'b) det_monad) ⇒
'e +'a ⇒ ('s, 'e + 'b) det_monad"
where
"lift f v ≡ case v of Inl e ⇒ throwError e
| Inr v' ⇒ f v'"
text ‹
The definition of @{term bind} in the exception monad (new
name ‹bindE›): the same as normal @{term bind}, but
the right-hand side is skipped if the left-hand side
produced an exception.
›
definition
bindE :: "('s, 'e + 'a) det_monad ⇒
('a ⇒ ('s, 'e + 'b) det_monad) ⇒
('s, 'e + 'b) det_monad" (infixl ">>=E" 60)
where
"bindE f g ≡ bind f (lift g)"
text ‹
Lifting a normal deterministic monad into the
exception monad is achieved by always returning its
result as normal result and never throwing an exception.
›
definition
liftE :: "('s,'a) det_monad ⇒ ('s, 'e+'a) det_monad"
where
"liftE f ≡ f >>= (λr. return (Inr r))"
text ‹
Since the underlying type and ‹return› function changed,
we need new definitions for when and unless:
›
definition
whenE :: "bool ⇒ ('s, 'e + unit) det_monad ⇒
('s, 'e + unit) det_monad"
where
"whenE P f ≡ if P then f else returnOk ()"
definition
unlessE :: "bool ⇒ ('s, 'e + unit) det_monad ⇒
('s, 'e + unit) det_monad"
where
"unlessE P f ≡ if P then returnOk () else f"
text ‹
Throwing an exception when the parameter is @{term None}, otherwise
returning @{term "v"} for @{term "Some v"}.
›
definition
throw_opt :: "'e ⇒ 'a option ⇒ ('s, 'e + 'a) det_monad" where
"throw_opt ex x ≡
case x of None ⇒ throwError ex | Some v ⇒ returnOk v"
subsection "Monad Laws for the Exception Monad"
text ‹More direct definition of @{const liftE}:›
lemma liftE_def2:
"liftE f = (λs. ((λ(v,s'). (Inr v, s')) (fst (f s)), snd (f s)))"
by (auto simp: Let_def liftE_def return_def split_def bind_def h1_def h2_def)
text ‹Left @{const returnOk} absorbtion over @{term bindE}:›
lemma returnOk_bindE [simp]: "(returnOk x >>=E f) = f x"
apply (unfold bindE_def returnOk_def)
apply (clarsimp simp: lift_def)
done
lemma lift_return [simp]:
"lift (return ∘ Inr) = return"
by (rule ext)
(simp add: lift_def throwError_def split: sum.splits)
text ‹Right @{const returnOk} absorbtion over @{term bindE}:›
lemma bindE_returnOk [simp]: "(m >>=E returnOk) = m"
by (simp add: bindE_def returnOk_def)
text ‹Associativity of @{const bindE}:›
lemma bindE_assoc:
"(m >>=E f) >>=E g = m >>=E (λx. f x >>=E g)"
apply (simp add: bindE_def bind_assoc)
apply (rule arg_cong [where f="λx. m >>= x"])
apply (rule ext)
apply (case_tac x, simp_all add: lift_def throwError_def)
done
text ‹@{const returnOk} could also be defined via @{const liftE}:›
lemma returnOk_liftE:
"returnOk x = liftE (return x)"
by (simp add: liftE_def returnOk_def)
text ‹Execution after throwing an exception is skipped:›
lemma throwError_bindE [simp]:
"(throwError E >>=E f) = throwError E"
by (simp add: bindE_def bind_def h1_def h2_def throwError_def lift_def return_def)
section "Syntax"
text ‹This section defines traditional Haskell-like do-syntax
for the state monad in Isabelle.›
subsection "Syntax for the Nondeterministic State Monad"
text ‹We use ‹K_bind› to syntactically indicate the
case where the return argument of the left side of a @{term bind}
is ignored›
definition
K_bind_def [iff]: "K_bind ≡ λx y. x"
nonterminal
dobinds and dobind and nobind
syntax
"_dobind" :: "[pttrn, 'a] => dobind" ("(_ ←/ _)" 10)
"" :: "dobind => dobinds" ("_")
"_nobind" :: "'a => dobind" ("_")
"_dobinds" :: "[dobind, dobinds] => dobinds" ("(_);//(_)")
"_do" :: "[dobinds, 'a] => 'a" ("(do ((_);//(_))//od)" 100)
translations
"_do (_dobinds b bs) e" == "_do b (_do bs e)"
"_do (_nobind b) e" == "b >>= (CONST K_bind e)"
"do x ← a; e od" == "a >>= (λx. e)"
text ‹Syntax examples:›
lemma "do x ← return 1;
return (2::nat);
return x
od =
return 1 >>=
(λx. return (2::nat) >>=
K_bind (return x))"
by (rule refl)
lemma "do x ← return 1;
return 2;
return x
od = return 1"
by simp
subsection "Syntax for the Exception Monad"
text ‹
Since the exception monad is a different type, we
need to syntactically distinguish it in the syntax.
We use ‹doE›/‹odE› for this, but can re-use
most of the productions from ‹do›/‹od›
above.
›
syntax
"_doE" :: "[dobinds, 'a] => 'a" ("(doE ((_);//(_))//odE)" 100)
translations
"_doE (_dobinds b bs) e" == "_doE b (_doE bs e)"
"_doE (_nobind b) e" == "b >>=E (CONST K_bind e)"
"doE x ← a; e odE" == "a >>=E (λx. e)"
text ‹Syntax examples:›
lemma "doE x ← returnOk 1;
returnOk (2::nat);
returnOk x
odE =
returnOk 1 >>=E
(λx. returnOk (2::nat) >>=E
K_bind (returnOk x))"
by (rule refl)
lemma "doE x ← returnOk 1;
returnOk 2;
returnOk x
odE = returnOk 1"
by simp
section "Library of Monadic Functions and Combinators"
text ‹Lifting a normal function into the monad type:›
definition
liftM :: "('a ⇒ 'b) ⇒ ('s,'a) det_monad ⇒ ('s, 'b) det_monad"
where
"liftM f m ≡ do x ← m; return (f x) od"
text ‹The same for the exception monad:›
definition
liftME :: "('a ⇒ 'b) ⇒ ('s,'e+'a) det_monad ⇒ ('s,'e+'b) det_monad"
where
"liftME f m ≡ doE x ← m; returnOk (f x) odE"
text ‹
Run a sequence of monads from left to right, ignoring return values.›
definition
sequence_x :: "('s, 'a) det_monad list ⇒ ('s, unit) det_monad"
where
"sequence_x xs ≡ foldr (λx y. x >>= (λ_. y)) xs (return ())"
text ‹
Map a monadic function over a list by applying it to each element
of the list from left to right, ignoring return values.
›
definition
mapM_x :: "('a ⇒ ('s,'b) det_monad) ⇒ 'a list ⇒ ('s, unit) det_monad"
where
"mapM_x f xs ≡ sequence_x (map f xs)"
text ‹
Map a monadic function with two parameters over two lists,
going through both lists simultaneously, left to right, ignoring
return values.
›
definition
zipWithM_x :: "('a ⇒ 'b ⇒ ('s,'c) det_monad) ⇒
'a list ⇒ 'b list ⇒ ('s, unit) det_monad"
where
"zipWithM_x f xs ys ≡ sequence_x (zipWith f xs ys)"
text ‹The same three functions as above, but returning a list of
return values instead of ‹unit››
definition
sequence :: "('s, 'a) det_monad list ⇒ ('s, 'a list) det_monad"
where
"sequence xs ≡ let mcons = (λp q. p >>= (λx. q >>= (λy. return (x#y))))
in foldr mcons xs (return [])"
definition
mapM :: "('a ⇒ ('s,'b) det_monad) ⇒ 'a list ⇒ ('s, 'b list) det_monad"
where
"mapM f xs ≡ sequence (map f xs)"
definition
zipWithM :: "('a ⇒ 'b ⇒ ('s,'c) det_monad) ⇒
'a list ⇒ 'b list ⇒ ('s, 'c list) det_monad"
where
"zipWithM f xs ys ≡ sequence (zipWith f xs ys)"
definition
foldM :: "('b ⇒ 'a ⇒ ('s, 'a) det_monad) ⇒ 'b list ⇒ 'a ⇒ ('s, 'a) det_monad"
where
"foldM m xs a ≡ foldr (λp q. q >>= m p) xs (return a) "
text ‹The sequence and map functions above for the exception monad,
with and without lists of return value›
definition
sequenceE_x :: "('s, 'e+'a) det_monad list ⇒ ('s, 'e+unit) det_monad"
where
"sequenceE_x xs ≡ foldr (λx y. doE _ ← x; y odE) xs (returnOk ())"
definition
mapME_x :: "('a ⇒ ('s,'e+'b) det_monad) ⇒ 'a list ⇒
('s,'e+unit) det_monad"
where
"mapME_x f xs ≡ sequenceE_x (map f xs)"
definition
sequenceE :: "('s, 'e+'a) det_monad list ⇒ ('s, 'e+'a list) det_monad"
where
"sequenceE xs ≡ let mcons = (λp q. p >>=E (λx. q >>=E (λy. returnOk (x#y))))
in foldr mcons xs (returnOk [])"
definition
mapME :: "('a ⇒ ('s,'e+'b) det_monad) ⇒ 'a list ⇒
('s,'e+'b list) det_monad"
where
"mapME f xs ≡ sequenceE (map f xs)"
text ‹Filtering a list using a monadic function as predicate:›
primrec
filterM :: "('a ⇒ ('s, bool) det_monad) ⇒ 'a list ⇒ ('s, 'a list) det_monad"
where
"filterM P [] = return []"
| "filterM P (x # xs) = do
b ← P x;
ys ← filterM P xs;
return (if b then (x # ys) else ys)
od"
section "Catching and Handling Exceptions"
text ‹
Turning an exception monad into a normal state monad
by catching and handling any potential exceptions:
›
definition
catch :: "('s, 'e + 'a) det_monad ⇒
('e ⇒ ('s, 'a) det_monad) ⇒
('s, 'a) det_monad" (infix "<catch>" 10)
where
"f <catch> handler ≡
do x ← f;
case x of
Inr b ⇒ return b
| Inl e ⇒ handler e
od"
text ‹
Handling exceptions, but staying in the exception monad.
The handler may throw a type of exceptions different from
the left side.
›
definition
handleE' :: "('s, 'e1 + 'a) det_monad ⇒
('e1 ⇒ ('s, 'e2 + 'a) det_monad) ⇒
('s, 'e2 + 'a) det_monad" (infix "<handle2>" 10)
where
"f <handle2> handler ≡
do
v ← f;
case v of
Inl e ⇒ handler e
| Inr v' ⇒ return (Inr v')
od"
text ‹
A type restriction of the above that is used more commonly in
practice: the exception handle (potentially) throws exception
of the same type as the left-hand side.
›
definition
handleE :: "('s, 'x + 'a) det_monad ⇒
('x ⇒ ('s, 'x + 'a) det_monad) ⇒
('s, 'x + 'a) det_monad" (infix "<handle>" 10)
where
"handleE ≡ handleE'"
text ‹
Handling exceptions, and additionally providing a continuation
if the left-hand side throws no exception:
›
definition
handle_elseE :: "('s, 'e + 'a) det_monad ⇒
('e ⇒ ('s, 'ee + 'b) det_monad) ⇒
('a ⇒ ('s, 'ee + 'b) det_monad) ⇒
('s, 'ee + 'b) det_monad"
("_ <handle> _ <else> _" 10)
where
"f <handle> handler <else> continue ≡
do v ← f;
case v of Inl e ⇒ handler e
| Inr v' ⇒ continue v'
od"
section "Hoare Logic"
subsection "Validity"
text ‹This section defines a Hoare logic for partial correctness for
the deterministic state monad as well as the exception monad.
The logic talks only about the behaviour part of the monad and ignores
the failure flag.
The logic is defined semantically. Rules work directly on the
validity predicate.
In the deterministic state monad, validity is a triple of precondition,
monad, and postcondition. The precondition is a function from state to
bool (a state predicate), the postcondition is a function from return value
to state to bool. A triple is valid if for all states that satisfy the
precondition, all result values and result states that are returned by
the monad satisfy the postcondition. Note that if the computation returns
the empty set, the triple is trivially valid. This means @{term "assert P"}
does not require us to prove that @{term P} holds, but rather allows us
to assume @{term P}! Proving non-failure is done via separate predicate and
calculus (see below).
›
definition
valid :: "('s ⇒ bool) ⇒ ('s,'a) det_monad ⇒ ('a ⇒ 's ⇒ bool) ⇒ bool"
("⦃_⦄/ _ /⦃_⦄")
where
"⦃P⦄ f ⦃Q⦄ ≡ ∀s. P s ⟶ (∀r s'. ((r,s') = fst (f s) ⟶ Q r s'))"
text ‹
Validity for the exception monad is similar and build on the standard
validity above. Instead of one postcondition, we have two: one for
normal and one for exceptional results.
›
definition
validE :: "('s ⇒ bool) ⇒ ('s, 'a + 'b) det_monad ⇒
('b ⇒ 's ⇒ bool) ⇒
('a ⇒ 's ⇒ bool) ⇒ bool"
("⦃_⦄/ _ /(⦃_⦄,/ ⦃_⦄)")
where
"⦃P⦄ f ⦃Q⦄,⦃E⦄ ≡ ⦃P⦄ f ⦃ λv s. case v of Inr r ⇒ Q r s | Inl e ⇒ E e s ⦄"
text ‹
The following two instantiations are convenient to separate reasoning
for exceptional and normal case.
›
definition
validE_R :: "('s ⇒ bool) ⇒ ('s, 'e + 'a) det_monad ⇒
('a ⇒ 's ⇒ bool) ⇒ bool"
("⦃_⦄/ _ /⦃_⦄, -")
where
"⦃P⦄ f ⦃Q⦄,- ≡ validE P f Q (λx y. True)"
definition
validE_E :: "('s ⇒ bool) ⇒ ('s, 'e + 'a) det_monad ⇒
('e ⇒ 's ⇒ bool) ⇒ bool"
("⦃_⦄/ _ /-, ⦃_⦄")
where
"⦃P⦄ f -,⦃Q⦄ ≡ validE P f (λx y. True) Q"
text ‹Abbreviations for trivial preconditions:›
abbreviation(input)
top :: "'a ⇒ bool" ("⊤")
where
"⊤ ≡ λ_. True"
abbreviation(input)
bottom :: "'a ⇒ bool" ("⊥")
where
"⊥ ≡ λ_. False"
text ‹Abbreviations for trivial postconditions (taking two arguments):›
abbreviation(input)
toptop :: "'a ⇒ 'b ⇒ bool" ("⊤⊤")
where
"⊤⊤ ≡ λ_ _. True"
abbreviation(input)
botbot :: "'a ⇒ 'b ⇒ bool" ("⊥⊥")
where
"⊥⊥ ≡ λ_ _. False"
text ‹
Lifting ‹∧› and ‹∨› over two arguments.
Lifting ‹∧› and ‹∨› over one argument is already
defined (written ‹and› and ‹or›).
›
definition
bipred_conj :: "('a ⇒ 'b ⇒ bool) ⇒ ('a ⇒ 'b ⇒ bool) ⇒ ('a ⇒ 'b ⇒ bool)"
(infixl "And" 96)
where
"bipred_conj P Q ≡ λx y. P x y ∧ Q x y"
definition
bipred_disj :: "('a ⇒ 'b ⇒ bool) ⇒ ('a ⇒ 'b ⇒ bool) ⇒ ('a ⇒ 'b ⇒ bool)"
(infixl "Or" 91)
where
"bipred_disj P Q ≡ λx y. P x y ∨ Q x y"
subsection "Determinism"
text ‹A monad of type ‹det_monad› is deterministic iff it
returns exactly one state and result and does not fail›
definition
det :: "('a,'s) det_monad ⇒ bool"
where
"det f ≡ ∀s. ∃r. f s = (r,False)"
text ‹A deterministic ‹det_monad› can be turned
into a normal state monad:›
definition
the_run_state :: "('s,'a) det_monad ⇒ 's ⇒ 'a × 's"
where
"the_run_state M ≡ λs. THE s'. fst (M s) = s'"
subsection "Non-Failure"
text ‹
With the failure flag, we can formulate non-failure separately
from validity. A monad ‹m› does not fail under precondition
‹P›, if for no start state in that precondition it sets
the failure flag.
›
definition
no_fail :: "('s ⇒ bool) ⇒ ('s,'a) det_monad ⇒ bool"
where
"no_fail P m ≡ ∀s. P s ⟶ ¬ (snd (m s))"
text ‹
It is often desired to prove non-failure and a Hoare triple
simultaneously, as the reasoning is often similar. The following
definitions allow such reasoning to take place.
›
definition
validNF ::"('s ⇒ bool) ⇒ ('s,'a) det_monad ⇒ ('a ⇒ 's ⇒ bool) ⇒ bool"
("⦃_⦄/ _ /⦃_⦄!")
where
"validNF P f Q ≡ valid P f Q ∧ no_fail P f"
definition
validE_NF :: "('s ⇒ bool) ⇒ ('s, 'a + 'b) det_monad ⇒
('b ⇒ 's ⇒ bool) ⇒
('a ⇒ 's ⇒ bool) ⇒ bool"
("⦃_⦄/ _ /(⦃_⦄,/ ⦃_⦄!)")
where
"validE_NF P f Q E ≡ validE P f Q E ∧ no_fail P f"
lemma validE_NF_alt_def:
"⦃ P ⦄ B ⦃ Q ⦄,⦃ E ⦄! = ⦃ P ⦄ B ⦃ λv s. case v of Inl e ⇒ E e s | Inr r ⇒ Q r s ⦄!"
by (clarsimp simp: validE_NF_def validE_def validNF_def)
section "Basic exception reasoning"
text ‹
The following predicates ‹no_throw› and ‹no_return› allow
reasoning that functions in the exception monad either do
no throw an exception or never return normally.
›
definition "no_throw P A ≡ ⦃ P ⦄ A ⦃ λ_ _. True ⦄,⦃ λ_ _. False ⦄"
definition "no_return P A ≡ ⦃ P ⦄ A ⦃λ_ _. False⦄,⦃λ_ _. True ⦄"
end