Re: [isabelle] Predicate Compiler fails with "No specification for ..."

Hi Florian,

code_pred's [inductify] option has its quirks and doesn't always work; in particular set comprehensions and set membership are tricky (code_pred was developed when "'a set" was a type abbreviation for "'a => bool"). What you can do is to do the inductification manually. In your example with the typedef, the following makes code_pred generate the right equations (of course, you should manually fill in the sorrys)

lemma P_intro [code_pred_intro]: "P n m"
  if "Rep_mynat n ≠ 0" "m = Abs_mynat (Rep_mynat n - 1)" sorry

code_pred (modes: i ⇒ i ⇒ bool, i ⇒ o ⇒ bool) P sorry

code_pred (modes: i ⇒ i ⇒ bool, i ⇒ o ⇒ bool) [inductify] t .

Now, however, export_code gives you a well-sortedness error that mynat isn't an instance of equal. But that's something you probably want to provide anyway; I'm actually surprised that the setup_lifting doesn't generate such an instance.

Hope this helps,

On 20.02.21 19:32, Florian Märkl wrote:
Hello Tobias,

in my larger project I could probably get rid of the typdefd types in
this specific predicate by just duplicating the code of what is called P
here, operating directly on the underying type of the typedef.

However after doing this, I end up with another issue that can be seen
in this example:

theory "Scratch"
   imports "HOL-Library.Transitive_Closure_Table" Main

definition P :: "nat ⇒ nat ⇒ bool" where "P a b ≡ a ∈ set [b]"
definition t where "t ≡ (P)^**"

(* Fails with:
exception TERM raised (line 80 of
   defining_const_of_introrule failed: Not a constant
   {x. x ∈ ?z} ≡ ?z
code_pred (modes: i ⇒ i ⇒ bool, i ⇒ o ⇒ bool) [inductify, show_steps,
show_intermediate_results] t .

(* The final goal *)
export_code t in SML


So it seems there is an issue with the ∈ operator now, which might be
possible to eliminate too (here by operating on the list [b], probably
similar in the real project) but this also brings me back to the
original problem: Is it somehow possible to use code_pred on a
definition like "t ≡ (P)^**" that does not use the "inductive" command
at the top level, without code_pred having to recurse through every
little definition that it contains?


On 2/20/21 6:07 PM, Tobias Nipkow wrote:
"No specification for Rep_mynat": I suspect the code generator just
cannot handle types specified by "typedef".


On 19/02/2021 15:03, Florian Märkl wrote:

I am trying to replicate something like discussed in

(the part of the JinjaThreads theory), i.e. generating code for a
reflexive transitive closure predicate, however the [inductify] option
that I have to use on code_pred currently seems to give me some

In essence, what I have in my project is something like this:

theory "Scratch"

    imports "HOL-Library.Transitive_Closure_Table" Main


typedef mynat = "{n :: nat. True}" by auto

setup_lifting type_definition_mynat

definition P :: "mynat ⇒ mynat ⇒ bool" where

    "P a b ≡ Rep_mynat a = Suc (Rep_mynat b)"

definition t where "t = (P)^**"

(* Fails with: "No specification for Rep_mynat" *)

code_pred (modes: i ⇒ i ⇒ bool, i ⇒ o ⇒ bool) [inductify] t .

(* The final goal *)

export_code t in SML


I have some executable predicate P over which I want to execute the
reflexive transitive closure t.

Now from what I understand (please correct me if I am wrong), the
[inductify] option here is necessary so code_pred understands that this
is actually an inductive predicate by digging into the definitions.

But that also means it will go even deeper and try to inspect Rep_mynat
for example, which fails.

What I intuitively would want it to do is to only unfold the t, but then
consider P an opaque constant.

I have seen that code_pred has quite a few more undocumented options,
but didn't find anything yet that did what I need. Is something like
this possible somehow or am I misunderstanding the inner workings of the
predicate compiler here?


This archive was generated by a fusion of Pipermail (Mailman edition) and MHonArc.