Operators

CEL operators are sugar for ordinary function calls. The checker resolves each operator to one or more overloads (e.g. + has overloads for int, uint, double, string, bytes, list); the runtime dispatches by overload id.

Arithmetic

Op	Types	Notes
+	int, uint, double, string, bytes, list<T>	concatenation for non-numeric kinds
-	int, uint, double, timestamp - timestamp → duration, timestamp - duration → timestamp	unary - for numeric
*	int, uint, double	overflow → error
/	int, uint, double	divide-by-zero → error
%	int, uint	divide-by-zero → error

Comparison

Op	Types
==, !=	any
<, <=, >, >=	bool, int, uint, double, string, bytes, duration, timestamp; cross-numeric mixes via double

CEL equality is symmetric across compatible numeric types: 1 == 1u == 1.0 is true. null == null is true. NaN is never equal to anything, including itself.

Logical

Op	Notes
&&	short-circuits; absorbs errors when result is false
||	short-circuits; absorbs errors when result is true
!	negation

The short-circuit / error-absorption rules are documented under Errors & unknowns.

Membership

Op	Notes
x in list	true if any element equals x
k in map	true if k is a key

Ternary

cond ? a : b

cond must be bool. Errors short-circuit per the spec table.

Indexing & selection

Form	Meaning
obj.field	attribute select
obj["k"]	index access (string key on map / dyn)
list[i]	index access (int key on list)
obj.?field	optional select — see Optionals
obj[?"k"]	optional index

Construction

Form	Meaning
[1, 2, 3]	list literal
{"a": 1, "b": 2}	map literal
pkg.Type{a: 1}	object construction (provider-mediated)

Operator overload ids

For function-style binding (e.g. when overriding equality on the runtime registry):

Operator	Overload ids
+	add_int_int, add_uint_uint, add_double_double, add_string_string, add_bytes_bytes, add_list_list, add_timestamp_duration, add_duration_duration, add_duration_timestamp
-	subtract_int_int, subtract_uint_uint, subtract_double_double, subtract_timestamp_timestamp, subtract_timestamp_duration, subtract_duration_duration, negate_int, negate_double
*	multiply_int_int, multiply_uint_uint, multiply_double_double
/	divide_int_int, divide_uint_uint, divide_double_double
%	modulo_int_int, modulo_uint_uint
==	equals (parametric over types)
!=	not_equals
<, <=, >, >=	less, less_equals, greater, greater_equals (cross-type and same-type variants)
&&, ||, !	logical_and, logical_or, logical_not
in	in_list, in_map
?:	conditional

These ids are stable across versions — feel free to bind alternative implementations on the runtime registry if you need custom semantics (see e.g. how CompiledProgram overrides equals to consult the ITypeProvider).

Precedence

From highest to lowest:

1. (), [], ., function call
2. !, unary -
3. *, /, %
4. +, -
5. <, <=, >, >=
6. ==, !=
7. in
8. &&
9. ||
10. ?:

This matches the cel-spec grammar. Use parentheses when the structure matters; CEL does not support custom operators.

See also

• Standard library — built-in functions that aren’t operators.
• Errors & unknowns — short-circuit table.