Generate rewrite relation

[tothtamas28]

Closes #4728

Generates an inductive type Rewrites : GeneratedTopCell -> GeneratedTopCell -> Prop where each constructor (except for the first one which expresses transitivity) encodes a K rewrite rule.

Each constructor has the following signature:

• An implicit binder for each free variable.
• An explicit binder for each function application which expresses definedness of the function over the arguments.
• An explicit binder for the requires clause.
• The type Rewrites <lhs> <rhs>.

[tothtamas28]

The Rewrites type definition generated for evm-semantics.llvm compiles in about 20 mins (after imports, patching ite and setting set_option maxHeartbeats 10000000).

Rewrite.txt

[JuanCoRo]

Where do the names of rewrite rules such as _43a65b8 come from?

[JuanCoRo]

Where do the names of rewrite rules such as _43a65b8 come from?

I guess from unnamed rules. But maybe it's better to have something like ruleN where N is the counter of how many rules there are. Otherwise proving things with this can get more convoluted.

A related question: Even if we named all rules in K definitions, will there be generated rules by the compiler that we cannot name?

[tothtamas28]

Where do the names of rewrite rules such as _43a65b8 come from?

It's a prefix of the UNIQUE-ID generated by the frontend for each rule.

But maybe it's better to have something like ruleN where N is the counter of how many rules there are.

I can make that change, but that'll make it significantly harder to manually trace back and forth between K and Lean source. Right now you can look up the source and UNIQUE-ID attributes from compiled.txt or definition.kore, which provides the mapping. And it's easier to search for a unique string in those files then to navigate to the n-th match of a regex identifying rule axioms.

A related question: Even if we named all rules in K definitions, will there be generated rules by the compiler that we cannot name?

The only frontend-generated rules I can think of are heating-cooling rules, but they seem to have a label. Still, it's possible there are unlabeled generated rules.

[JuanCoRo]

It's a prefix of the UNIQUE-ID generated by the frontend for each rule.

Is it correct that the frontend will generate such UNIQUE-ID if the rule is unnamed? If so, perhaps we should name all rules for a better Lean 4 user experience.

that'll make it significantly harder to manually trace back and forth between K and Lean source

Makes sense! Let's leave it like this then.

[tothtamas28]

Is it correct that the frontend will generate such UNIQUE-ID if the rule is unnamed?

Yes, the UNIQUE-ID is generated for all rules.

perhaps we should name all rules for a better Lean 4 user experience.

That would be ideal, I agree. Similarly, all symbol productions should have the symbol attribute to avoid KORE symbol names like

_________EVM_InternalOp_CallOp_Int_Int_Int_Int_Int_Int_Int