eytans / thesy Goto Github PK

Types with only one constructor (such as tuples) are just wrappers for other data therefore they should always be expanded by the constructor.
This requires a few technical details to work:

1. During datatype initialization separate datatypes to later be deconstructed
2. During rewriting recognize which parameters are of said types
3. When deconstructing we are essentially creating existentials, so they should be wrapped with a function call to make them unique in regard to the original term or have a unique identifier

See the example victor provided (there should be a branch for it) to start

Update the parsing of terms in thesy_parser.rs at fn collect_rule(l: &mut Vec) to parse relevant searchers.
Look at searchers in eggstentions

Currently, the runtime is greatly affected by get_conjectures, which might take more than half the runtime.
It seems that using Extractor from egg is inefficient for our case.
Implementation of get_conjectures should have iterations to the depth of known terms (no need to find terms deeper than those generated in the enumeration), although deeper terms exist.

Consider a term of addition, 2 + 2. the depth of the term is 3, 1 for the + and 2 for the number (i.e. S (S 0)).
After rewriting, we get a depth 4 term, 4.
We can ignore most of the graph and focus only on Eclasses that exist inside the id_examples data structure.
This way we can prevent unnecessary computations.

The algorithm should iterate through the nodes in the EGraph building terms:

1. Create a vector for each EClass, for each node, with its dependencies (i.e. other EClasses).

2. Remember all edges without dependencies.

3. For each class remember nodes depending on it.

4. Iteratively go through all edges with no remaining dependencies:

   • "Free up" all dependant nodes
   • If a node is fully freed add an EClass to the next iteration
   • Keep for each class its minimal representation

Also verify that other searchers are fixed

We want to be able to participate in first order solver competitions.
Some of the features might not be supported in TheSy, we should open issues on each and see which are technical limitations and which require further research.

http://www.tptp.org/

FilteringSearcher is filtering cases where with the addition of a color the predicate will pass. We should fix this with either a new colored_filter, or with changing the existing filter function