philzook58 / egglog0

Datalog + Egg = Good

Defunct repo: Please see https://github.com/egraphs-good/egglog

egglog0

Using the egg library with a file format and semantics similar to datalog.

Short talk proposal paper for EGRAPHS 2022 https://github.com/philzook58/egglog0-talk/blob/main/out.pdf

Explanatory blog posts:

• https://www.philipzucker.com/egglog-checkpoint/
• https://www.philipzucker.com/egglog2-monic/

Try It Online!!!

http://www.philipzucker.com/egglog/

Building

To run on a file locally: cargo run --release tests/examples.pl

To build the wasm library: wasm-pack build --target web

Note: I started modifying egg a bit. I exposed the Subst datatypes field.

Ideas

• MultiApplier could be useful / efficient
• Using Conditional Equals could be useful if all variables known
• But really getting patterns to compile with substituion pieces considered known subsumes this optimization I think in modern egg with yihong's optimization.
• Sanity checks that needed variables exist would be good. when it does crash it names rules, so that's something.
• May want to run Runner multiple times since it may not get restarted. Currently I have that vec![0] hack
• _ for dummy variables
• The ability to check to see if something is in the egraph.
• graphviz dumping the egraph. graphviz wasm?
• harrop formula
• merge_subst that doesn't copy?
• Give rules names. Keep a hash table of them?
• Queries with variables
• Queries should be conjunctions
• a REPL would be sweet. especially if we have higher order rules, we could watch the database, add queries
• termination based on the query condition
• side effectful searchers and appliers (printing mostly), functions.
• Astsize with weighting? Does that get me anywhere?
• infix operators
• better printers
• rewrite/proof files that allow intermediate queriess. set of support?
• cli
• smtlib subset (forall (a b ) (= (f a) (g c)) ) ! :pattern) or horn cluase style.
• vaguely ML/coq style synax
• tptp syntax?
• push pop directives instead of clear.
• only allow stuff that compresses the egraph? Appliers that do not add terms to the egraph or only add a couple? Or keeps counts.
• directives to changes egraph params. or flags?
• Macros/simplification stage?
• typed symbollang - would this even be an optimization?
• defunctionalization of lambdas. lambda-egglog
• backchain until stumped? depth limitted backchain?
• hashlog - experiment with same thing but on hashcons instead of egraph. Easier to understand semi naive?
• epeg extraction
• faster multipattern via compilation
• integerate analysis?
• gensym
• serialize the egraph
• negation checks. nonlogical
• cli options to the runner

Tests

cram tests/cram/*.t -i