Performance degradations v0.1.6 -> v0.2.5

[patham9]

Describe the bug

MeTTa v0.2.5 is significantly slower (more than 10x) than v0.1.6 in some examples.
Interestingly in many other benchmark codes I had it has improved performance by a factor of 2-3, but in the cases where it degraded it is usually a way higher factor, which could hint to a deeper issue that got introduced.

The following is a quite small example which I hope will be helpful to investigate:

(= (Add $x Z) $x)
(= (Add $x (S $y)) (Add (S $x) $y))
(= (Add (S $x) $y) (S (Add $x $y)))

!(Add (S (S (S Z)))
      (S (S (S Z))))

To Reproduce
Steps to reproduce the behavior:
Running the example with the Rust MeTTa interpreter.

Expected behavior
Similar solving time as v0.1.6.

Actual behavior
About 10-15x slower.

Additional context

• MeTTa version: v0.2.5 compared with v0.1.6

[patham9]

I was told this comes from the interpreter now handling variable bindings more correctly, so the comparison might not be fair.

[vsbogd]

@patham9 yes, one of the reasons of the slowness is that interpreter started handling dependencies among variables more correctly and using minimal metta as a basis of the interpreter introduced much more variables into the equation. Thus I would say current performance issues of the hyperon-experimental are more related to the variables handing than to the speed of the atomspace queries.

[vsbogd]

This is why I am thinking about introducing WAM memory model to handle variable bindings into hyperon-experimental.