Performance improvements to CEL rule evaluation

Three improvements to hot-path validation, inspired by protovalidate-go.

Guard `_validate_cel` when `self._cel` is empty (bufbuild/protovalidate-go#261)

Adds an early return in `CelRules._validate_cel` when there are no CEL
runners, skipping activation dict creation and `datetime.now()` entirely.
Also guards the `_msg_to_cel` call in `MessageRules.validate()` so the
message-to-CEL conversion is skipped when there are no CEL rules to run.
Benchmark (required-only field, 0 CEL runners): -37%.

Skip `now` computation when unused (bufbuild/protovalidate-go#289)

Adds `_uses_now` to `CelRules`, set at compile time in `add_rule` by
checking whether `"now"` appears in the expression string. Only
`timestamp.gt_now`, `timestamp.lt_now`, `timestamp.within`, and custom
expressions referencing `now` will call `datetime.datetime.now()`.
Benchmarks: int32.gt (5 CEL runners) -25%, timestamp.gt_now -6%.

Early-exit loop in `cel_unique` (bufbuild/protovalidate-go#289)

Replaces `len(val) == len(set(val))` with a loop that returns `False` on
the first duplicate, avoiding building the full set unnecessarily.

Author: stefanvanburen

protovalidate/internal/extra_func.py

@@ -328,7 +328,12 @@ def cel_unique(val: celtypes.Value) -> celpy.Result:
     if not isinstance(val, celtypes.ListType | list):
         msg = "invalid argument, expected list"
         raise celpy.CELEvalError(msg)
-    return celtypes.BoolType(len(val) == len(set(val)))
+    seen: set[celtypes.Value] = set()
+    for item in val:
+        if item in seen:
+            return celtypes.BoolType(False)  # noqa: FBT003
+        seen.add(item)
+    return celtypes.BoolType(True)  # noqa: FBT003
 
 
 class Ipv4:

protovalidate/internal/rules.py

@@ -342,6 +342,7 @@ class CelRules(Rules):
     _cel: list[CelRunner]
     _rules: message.Message | None = None
     _rules_cel: celtypes.Value | None = None
+    _uses_now: bool = False
 
     def __init__(self, rules: message.Message | None):
         self._cel = []
@@ -357,11 +358,14 @@ def _validate_cel(
         this_cel: celtypes.Value | None = None,
         for_key: bool = False,
     ):
+        if not self._cel:
+            return
         activation: dict[str, celtypes.Value] = {}
         if this_cel is not None:
             activation["this"] = this_cel
         activation["rules"] = self._rules_cel
-        activation["now"] = celtypes.TimestampType(datetime.datetime.now(tz=datetime.timezone.utc))
+        if self._uses_now:
+            activation["now"] = celtypes.TimestampType(datetime.datetime.now(tz=datetime.timezone.utc))
         for cel in self._cel:
             activation["rule"] = cel.rule_cel
             result = cel.runner.evaluate(activation)
@@ -409,6 +413,8 @@ def add_rule(
             rules = validate_pb2.Rule()
             rules.id = expression
             rules.expression = expression
+        if "now" in rules.expression:
+            self._uses_now = True
         ast = env.compile(rules.expression)
         prog = env.program(ast, functions=funcs)
         rule_value = None
@@ -463,9 +469,10 @@ def __init__(self, rules: message.Message | None, desc: descriptor.Descriptor):
         self._desc = desc
 
     def validate(self, ctx: RuleContext, message: message.Message):
-        self._validate_cel(ctx, this_cel=_msg_to_cel(message))
-        if ctx.done:
-            return
+        if self._cel:
+            self._validate_cel(ctx, this_cel=_msg_to_cel(message))
+            if ctx.done:
+                return
         for oneof in self._oneofs:
             oneof.validate(ctx, message)
             if ctx.done: