How to define an inductive predicate inside locale? - isabelle

Here is an example of a simple locale:
locale test =
fixes test_less_eq :: "'a ⇒ 'a ⇒ bool"
begin
inductive test_eq where
"test_less_eq x y ⟹ test_less_eq y x ⟹ test_eq x y"
end
It defines inductive test_eq. It can be defined using definition, but I need it to be an inductive predicate.
Then I define a trivial interpretation of the locale and try to use it:
interpretation interp: test "op <" .
inductive some_pred where
"interp.test_eq x y ⟹
some_pred x y"
code_pred [show_modes] some_pred .
The problem is that I get the following error for code_pred:
Type mismatch of predicate test.test_eq (trying to match ?'a
⇒ ?'a ⇒ bool and ('a ⇒ 'a ⇒ bool)
⇒ 'a ⇒ 'a ⇒ bool) in ?x1 < ?y1 ⟹
?y1 < ?x1 ⟹ interp.test_eq ?x1 ?y1
What is a cause of the error and how to fix it?

The predicate compiler has never been localized, i.e., it cannot directly deal with predicates that are defined inside a locale. There are two ways to make this work nevertheless.
Either, use global_interpretation with a defines clause to introduce a new constant for the predicate (plain interpretation only introduces an abbreviation). Then, you also have to re-declare the introduction rules to code_pred and prove the corresponding elimination rule.
global_interpretation interp: test "op <"
defines interp_test_eq = interp.test_eq .
declare interp.test_eq.intros[code_pred_intro]
code_pred interp_test_eq by(rule interp.test_eq.cases)
Or, leave the interpretation as is and re-declare the introduction rules of the internal constant to which the definition in the locale is mapped. This is <locale_name>.<predicate_name>, i.e., test.test_eq in your case. This works only if your locale has no assumption.
declare test.test_eq.intros[code_pred_intro]
code_pred test.test_eq by(rule test.test_eq.cases)

Related

Isabelle Code Generation and Linear Order

I am trying to use the export_code tool for the following definition:
definition set_to_list :: "('a×'a) set ⇒ ('a×'a) list"
where "set_to_list A = (SOME L. set L = A)"
This is not working due to missing code equations for Eps. Now I discovered that there is also a definition:
definition sorted_list_of_set :: "'a set ⇒ 'a list" where
"sorted_list_of_set = folding.F insort []"
However, I am not capable of asserting that ('a ×'a) is a linear order (which would be fine for me, e.g. first comparing the first element and then the second). Can someone help me to either fix my own definition or using the existing one?
To use sorted_list_of_set you can implement the type class linorder for product types:
instantiation prod :: (linorder,linorder) linorder begin
definition "less_eq_prod ≡ λ(x1,x2) (y1,y2). x1<y1 ∨ x1=y1 ∧ x2≤y2"
definition "less_prod ≡ λ(x1,x2) (y1,y2). x1<y1 ∨ x1=y1 ∧ x2<y2"
instance by standard (auto simp add: less_eq_prod_def less_prod_def)
end
You can also import "HOL-Library.Product_Lexorder" from the standard library, which includes a similar definition.
Then you can define set_to_list if you restrict the type parameter to implement linorder:
definition set_to_list :: "('a::linorder×'a) set ⇒ ('a×'a) list"
where "set_to_list A = sorted_list_of_set A"

Is it a good idea to extend standard types and operations?

I need to defined an extended bool type (ebool = bool ∪ {⊥}) and a set of operations for the type (conjunction, etc.).
Here is the theory:
theory EboolTest
imports Main "~~/src/HOL/Library/Adhoc_Overloading"
begin
notation
bot ("⊥")
declare [[coercion_enabled]]
typedef ebool = "UNIV :: bool option set" ..
definition ebool :: "bool ⇒ ebool" where
"ebool b = Abs_ebool (Some b)"
declare [[coercion "ebool :: bool ⇒ ebool"]]
instantiation ebool :: bot
begin
definition "⊥ ≡ Abs_ebool None"
instance ..
end
free_constructors case_ebool for
ebool
| "⊥ :: ebool"
apply (metis Rep_ebool_inverse bot_ebool_def ebool_def not_Some_eq)
apply (smt Abs_ebool_inverse ebool_def iso_tuple_UNIV_I option.inject)
by (simp add: Abs_ebool_inject bot_ebool_def ebool_def)
lemmas ebool2_cases = ebool.exhaust[case_product ebool.exhaust]
lemmas ebool3_cases = ebool.exhaust[case_product ebool.exhaust ebool.exhaust]
fun ebool_and :: "ebool ⇒ ebool ⇒ ebool" (infixr "∧" 35) where
"ebool_and a b = ebool (HOL.conj a b)"
| "ebool_and False _ = False"
| "ebool_and _ False = False"
| "ebool_and ⊥ _ = ⊥"
| "ebool_and _ ⊥ = ⊥"
no_notation HOL.conj (infixr "∧" 35)
consts "(∧)" :: "'a ⇒ 'a ⇒ 'a"
adhoc_overloading "(∧)" HOL.conj
adhoc_overloading "(∧)" ebool_and
end
The following works fine:
value "True ∧ (False::ebool)"
value "True ∧ ⊥"
But the following returns ebool, but I expect to see bool:
value "True ∧ False"
It seems my approach is bad. Could you suggest a better approach? Maybe I it's not good to overload standard operations at all?
First of all, I'm a little surprised that this:
consts "(∧)" :: "'a ⇒ 'a ⇒ 'a"
works at all. It sounds like a bug, because the (...) notation is reserved for the system. (To be fair, it does print a warning, and unless you know exactly what you're doing, it's unwise to ignore them.)
But to circle back to your actual problem, I don't think you should use adhoc_overloading to overload syntax that would otherwise be provided by Main.
There are alternatives. For example, you could use a bold version. This is used in this theory.
Or you could use another symbol, like &&.
As an addendum: I believe there might be weird interactions between coercions and adhoc overloading. Both tools are fine by themselves, but watch out if they interact with each other.

How to use classes or locales?

I'm trying to define a generic operations for a programming language:
type_synonym vname = "string"
type_synonym 'a env = "vname ⇒ 'a option"
locale language =
fixes big_step :: "'exp × 'val env ⇒ 'val ⇒ bool" (infix "⇒" 55)
fixes typing :: "'type env ⇒ 'exp ⇒ 'type ⇒ bool" ("(1_/ ⊢/ (_ :/ _))" [50,0,50] 50)
For example this is a particular language:
datatype foo_exp =
FooBConst bool |
FooLet vname foo_exp foo_exp |
FooVar vname |
FooAnd foo_exp foo_exp
datatype foo_val = FooBValue bool | FooIValue int
type_synonym foo_env = "foo_val env"
datatype foo_type = FooBType | FooIType
type_synonym foo_tenv = "foo_type env"
inductive foo_big_step :: "foo_exp × foo_env ⇒ foo_val ⇒ bool"
inductive foo_typing :: "foo_tenv ⇒ foo_exp ⇒ foo_type ⇒ bool"
How to make it an instance of language locale?
Is it possible to use same notation (⇒ and _ ⊢ _ : _) for different languages in one theory? Could this notation be polymorphic?
To specialize the parameters of a locale, you need to do an interpretation as in
interpretation foo: language foo_big_step foo_typing .
This will generate an abbreviation foo.f for every definition f in the locale language specialised to foo_big_step and foo_typing and every theorem thm of language becomes specialised to foo.thm. The mixfix syntax annotations of parameters and all constants in the locale will not be inherited.
Type classes cannot be used in this context because your locale depends on multiple type variables and type classes in Isabelle support only exactly one type variable.
If you want to use some kind of polymorphic notation for the big-step semantics and type judgements, Adhoc_Overloading might work, provided that Isabelle's parser can statically resolve the overloading uniquely. Here's how this might work:
theory Language imports Main "~~/src/Tools/Adhoc_Overloading" begin
type_synonym 'a env = "vname ⇒ 'a option"
consts
big_step :: "'exp × 'val env ⇒ 'val ⇒ bool" (infix "⇒" 55)
typing :: "'type env ⇒ 'exp ⇒ 'type ⇒ bool" ("(1_/ ⊢/ (_ :/ _))" [50,0,50] 50)
locale language =
fixes big_step :: "'exp × 'val env ⇒ 'val ⇒ bool"
fixes typing :: "'type env ⇒ 'exp ⇒ 'type ⇒ bool"
begin
adhoc_overloading Language.big_step big_step
adhoc_overloading Language.typing typing
end
After the interpretation, you have to register foo's semantics and type judgement constants foo_big_step and foo_typing for adhoc overloading with the syntactic constants big_step and typing again.
interpretation foo: language foo_big_step foo_typing .
adhoc_overloading Language.big_step foo_big_step
adhoc_overloading Language.typing foo_typing
So when you write
term "(x :: foo_exp, E) ⇒ v"
thereafter, Isabelle's parser will figure out by the types that this refers to foo_big_step, and inside the locale Language, term "(x :: 'exp, E) ⇒ v" is resolved to the locale parameter big_step.
This should also work for multiple interpretations of the locale Language provided that the types are sufficient to uniquely resolve the overloading. If not, you'll get error messages, which are not always easy to understand.

Working with generic definitions in Isabelle

I am working with limits and I am unable to prove the following
definition func :: "real ⇒ real"
where "func = real"
lemma "(λh. (func (x+h))) -- 0 --> (func (x))"
unfolding func_def
apply (auto intro!: tendsto_eq_intros)
However if I replace the definition of func to
definition func :: "real ⇒ real"
where "func x = x"
the lemma is solved.
How can I solve this lemma when working with generic definitions?
I believe, here the problem is that the function real has just a generic (overloaded) syntax, i.e., real :: 'a => real, but it is not necessarily defined for all possible types 'a. This is easily seen when using find_theorems: when searching for lemmas on real :: nat => real, you get plenty of results whereas searching for real :: real => real doesn't give you a single result.
find_theorems "real :: real => real"
find_theorems "real :: nat => real"
Consequently, you can not even prove a simple lemma like func x = x, since it is not specified that real :: real => real really is the identity function.

How do I abstractly work with theorems and terms using locale assumptions and variables in ML?

Consider the following locale definition:
locale my_locale =
fixes a :: nat
assumes "a > 0"
begin
definition "f n ≡ a + n"
lemma f_pos: "f x > 0"
sorry
end
In Isar, if I attempt to work with the definition of f or the lemma f_pos, the locale assumptions and fixed variables are hidden from me. For example, thm f_def f_pos returns:
f ?n ≡ a + ?n
0 < f ?x
as expected.
If, however, I try to reason about these terms in ML, the "hidden" fixed variables are suddenly exposed. ML {* #{thm f_def} |> prop_of *}, for instance, returns:
Const ("==", "nat ⇒ nat ⇒ prop") $
(Const ("TestSimple.my_locale.f", "nat ⇒ nat ⇒ nat") $
Free ("a", "nat") $ Var (("n", 0), "nat")) $
(Const ("Groups.plus_class.plus", "nat ⇒ nat ⇒ nat") $
Free ("a", "nat") $ Var (("n", 0), "nat"))
where the fixed variable a becomes a parameter to the function f.
Is there a way to be able to work inside a locales in ML so that I am not exposed to such locale variables?
It appears that the version of f which doesn't have the parameter a is simply an abbreviation generated by the locale command. In particular, typing print_abbrevs shows:
local.f ≡ My_Theory.my_locale.f a
This means that from the user's perspective, f doesn't appear to have any locale parameters, as they are hidden behind the abbreviation. Behind the scenes, however, f will always have the locale parameter attached to it, and thus ML code must be coded to explicitly handle it.

Resources