This is the working repo for egg-smol.
See this presentation for background information:
https://youtu.be/dbgZJyw3hnk?t=3983
with this paper to go along:
https://effect.systems/doc/pldi-2022-egraphs/abstract.pdf
Also see papers about egglog.
See also the Python binding, which provides a bit more documentation: https://egg-smol-python.readthedocs.io/en/latest/
There is a Zulip chat about egg-smol here: https://egraphs.zulipchat.com/#narrow/stream/328979-Implementation/topic/Eggsmol
apt-get install make cargo
make all
cargo run [-f fact-path] [-naive] <files.egg>
or just
cargo run
for the REPL.
There is a VS Code extension in the vscode folder. Install using 'Install from VSIX...' in the three-dot menu of the extensions tab and pick vscode/vscode/eggsmol.vsix.
If you want to hack on the VS Code extension, install nodejs, and make your changes in the files in the vscode/eggsmol-1.0.0 folder.
Then run
code vscode/eggsmol-1.0.0
and use F5 to run the extension in a new window. When satisfied, then install VSCE if you do not already have it:
npm install -g @vscode/vsce
Run vsce package in the vscode/eggsmol-1.0.0 folder to reconstruct the .vsix file and install it manually.
The syntax of the .egg files is defined in src/ast/parse.lalrpop.
Declare a datatype with this syntax:
( datatype <name:Ident> <variants:(Variant)*> )
where variants are:
( <name:Ident> <types:(Type)*> <cost:Cost> )
Example:
(datatype Math
(Num i64)
(Var String)
(Add Math Math)
(Mul Math Math))
defines a simple Math datatype with variants for numbers, named variables, addition and multiplication.
Datatypes are also known as algebraic data types, tagged unions and sum types.
( function <name:Ident> <schema:Schema> <cost:Cost>
(:on_merge <List<Action>>)?
(:merge <Expr>)?
(:default <Expr>)?
Defines a named function with a type schema, an optional integer cost, and an optional :on_merge or :merge expression, which can refer to old and new values. You can also provide a default value using :default.
Example:
(function add (Math Math) Math)
defines a function add which adds two Math datatypes and gives a Math as the result.
Functions are basically lookup tables from input tuples to the output. They can be considered as a kind of database table.
Explicit function values can be defined using set:
(function Fib (i64) i64)
(set (Fib 0) 0)
(set (Fib 1) 1)
You can extract the value of specific points of the function using extract:
(extract (Fib 1))
If you define a :merge expression, you can update specific values in the function, and the function relation will be updated using the merge expression:
(function KeepMax (i64) i64 :merge (max new old)); when updated, keep the biggest value
(set (KeepMax 0) 0)
(set (KeepMax 1) 1)
(set (KeepMax 1) 2) ; we redefine 1 to be 2
(set (KeepMax 1) 0) ; this does not change since we use max
(extract (KeepMax 1)) ; this is 2
The relation command defines a named function which returns the Unit type.
( relation <name:Ident> <types:List<Type>> )
Thus (relation <name> <args>) is equivalent to (function <name> <args> Unit).
Example:
(relation path (i64 i64))
(relation edge (i64 i64))
defines a path and an edge relation between two i64s.
(edge 1 2)
(edge 2 3)
inserts two edges into the store for the edge function. If your function is relation between the inputs, use relation and the above syntax to define the relations, since there is no syntax to define a unit value using set.
( define <name:Ident> <expr:Expr> <cost:Cost> )
defines a named value. This is the same as a 0-arity function with a given, singular value.
Example:
(define one 1)
(define two 2)
(define three (+ one two))
(extract three); extracts 3 as a i64
( rule <body:List<Fact>> <head:List<Action>> )
defines a rule, which matches a list of facts, and runs a bunch of actions. It is useful to maintain invariants and inductive definitions.
Example:
(rule ((edge x y))
((path x y)))
(rule ((path x y) (edge y z))
((path x z)))
These rules maintains path relations for a graph: If there is an edge from x to y, there is also a path from x to y. Transitivity is handled by the second rule: If there is a path from x to y and there is an edge from y to z, there is also a path from x and z.
( extract <variants:(:variants <UNum>)?> <e:Expr> )
where variants are:
( <name:Ident> <types:(Type)*> <cost:Cost> )
The extract queries the store to find the cheapest values matching the expression.
( rewrite <lhs:Expr> <rhs:Expr>
<conditions:(:when <List<Fact>>)?>
)
defines a rule which matches the lhs expressions, and rewrites them to the rhs expression. It is possible to guard the rewrite with a condition that has to be satisfied before the rule applies.
( birewrite <lhs:Expr> <rhs:Expr>
<conditions:(:when <List<Fact>>)?>
)
does the same, but where both directions apply.
Example:
(rewrite (Add a b)
(Add b a))
declares a rule that a Add variant is commutative.
(birewrite (* (* a b) c) (* a (* b c)))
declares a rule that multiplication is associative in both directions.
( check <Fact> )
( fail <Command> )
This evaluates the fact and checks that it is true.
Example:
(check (= (+ 1 2) 3))
(check (<= 0 3) (>= 3 0))
(fail (check (= 1 2)))
prints
[INFO ] Checked.
[INFO ] Checked.
[ERROR] Check failed
[INFO ] Command failed as expected.
( sort <name:Ident> ( <head:Ident> <tail:(Expr)*> ) )
( run <limit:UNum> <until:(:until <Fact>)?> ) ; evaluate rules N steps or until a condition is met
( clear-rules ) ; clear out all rules and rewrites
( clear ) ; clear data from the functions, not the function tables themselves
( query <List<Fact>> )
( push <UNum?> ) ; saves the state of the database on the stack
( pop <UNum?> ) ; restores the state of the database on the stack
( print <sym:Ident> <n:UNum?> ) ; print the value of an id
( print-size <sym:Ident> )
( input <name:Ident> <file:String> )
( output <file:String> <exprs:Expr+> ) ; Appends the expression to a file
( include <file:String> )
( add-ruleset <id:String> ) ; Saves all rules as a ruleset with a given name (EXPERIMENTAL)
( load-ruleset <id:String> ) ; Add the rules from a ruleset previously added (EXPERIMENTAL)
( calc ( <idents:IdentSort*> ) <exprs:Expr+> )
where sorts are:
( <ident:Ident> <sort:Type> )
( set ( <f: Ident> <args:Expr*> ) <v:Expr> )
( delete ( <f: Ident> <args:Expr*> ) )
( union <e1:Expr> <e2:Expr> )
( panic <msg:String> )
( let <name:Ident> <expr:Expr> )
The underlying data structure maintained by egg-smol is an e-graph. That means that specific values can be unified to be equivalent. To extract a value, use extract and it will extract the cheapest option according to the costs.
(datatype Math (Num i64))
(union (Num 1) (Num 2)); Define that Num 1 and Num 2 are equivalent
(extract (Num 1)); Extracts Num 1
(extract (Num 2)); Extracts Num 1
Union only works on variants, not sorts.
[ <Ident> ]
( = <mut es:Expr+> <e:Expr> )
<Expr>
These are conditions used in check and other commands. There is no boolean type in egg-smol. Instead, boolean are modelled morally as Option<Unit>, so if something is true, it is Some<()>. If something is false, it does not match and is None.
integer
string
identifier
call: ( <head:Ident> <tail:(Expr)*> )
Signed 64-bit integers supporting these primitives:
+ - * / % ; arithmetic
& | ^ << >> not-i64 ; bit-wise operations
< > <= >= ; comparisons
min max
64-bit floating point numbers supporting these primitives:
+ - * / % ; arithmetic
< > <= >= ; comparisons
min max
A map from a key type to a value type supporting these primitives:
empty
insert
get
not-contains
contains
set-union
set-diff
set-intersect
map-remove
Rational numbers (fractions) with 64-bit precision for numerator and denominator with these primitives:
+ - * / ; arithmetic
min max neg abs floor ceil round
rational ; construct from a numerator and denominator
pow log sqrt
< > <= >= ; comparisons
These primitives are only defined when the result itself is a pure rational.
Use double quotes to get a quote: "Foo "" Bar" is Foo " Bar.
No primitives defined.