César ValadezandClaude Sonnet 4.6
fcd53df552
Replace dynamicprompts.WildcardManager with a native implementation in src/wildcards.py that loads .txt, .yaml, and .json wildcard files from one or more directories using fnmatch pattern matching. EvaluationContext now types wildcard_manager as a WildcardProvider Protocol rather than the concrete class. Remove dynamicprompts from dependencies; add pyyaml>=6.0 and types-pyyaml as its replacement. Add explicit_package_bases = true to mypy config to resolve duplicate module path ambiguity from the root __init__.py. Update README and CLAUDE to reflect the native parser architecture and removal of the Jinja2 node. Remove completed plan documents. Ignore user-specific yaml wildcard files via .gitignore. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
101 lines
3.2 KiB
Python
101 lines
3.2 KiB
Python
import random
|
|
import pytest
|
|
from pathlib import Path
|
|
|
|
from src.evaluator.context import EvaluationContext
|
|
from src.evaluator.random_eval import evaluate
|
|
from src.parser.parser import parse
|
|
from src.wildcards import WildcardManager
|
|
|
|
|
|
@pytest.fixture
|
|
def ctx(wildcards_path: Path) -> EvaluationContext:
|
|
return EvaluationContext(
|
|
rng=random.Random(42),
|
|
wildcard_manager=WildcardManager(paths=[wildcards_path]),
|
|
)
|
|
|
|
|
|
def test_plain_text(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("hello world"), ctx)
|
|
assert result == "hello world"
|
|
|
|
|
|
def test_variant_picks_one(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("{cat|dog}"), ctx)
|
|
assert result in {"cat", "dog"}
|
|
|
|
|
|
def test_same_seed_reproducible() -> None:
|
|
t = parse("{cat|dog|bird}")
|
|
r1 = evaluate(t, EvaluationContext(rng=random.Random(42), wildcard_manager=WildcardManager()))
|
|
r2 = evaluate(t, EvaluationContext(rng=random.Random(42), wildcard_manager=WildcardManager()))
|
|
assert r1 == r2
|
|
|
|
|
|
def test_weighted_variant() -> None:
|
|
# Weight 0 means never picked
|
|
results = {
|
|
evaluate(parse("{1000::always|0::never}"), EvaluationContext(rng=random.Random(i), wildcard_manager=WildcardManager()))
|
|
for i in range(20)
|
|
}
|
|
assert "always" in results
|
|
assert "never" not in results
|
|
|
|
|
|
def test_multi_pick(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("{2$$a|b|c}"), ctx)
|
|
parts = result.split(", ")
|
|
assert len(parts) == 2
|
|
assert len(set(parts)) == 2 # no duplicates
|
|
|
|
|
|
def test_multi_pick_custom_separator(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("{2$$ and $$a|b|c}"), ctx)
|
|
assert " and " in result
|
|
|
|
|
|
def test_wildcard_resolution(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("__animals__"), ctx)
|
|
assert result in {"cat", "dog", "bird"}
|
|
|
|
|
|
def test_wildcard_in_template(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("A __animals__ running"), ctx)
|
|
assert result.startswith("A ")
|
|
assert result.endswith(" running")
|
|
|
|
|
|
def test_variable_immediate(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("${x=!{a|b}} ${x} ${x}"), ctx)
|
|
words = [p for p in result.split() if p]
|
|
assert len(words) == 2 # two uses, same value
|
|
assert words[0] == words[1]
|
|
|
|
|
|
def test_variable_non_immediate() -> None:
|
|
# Non-immediate: re-evaluated each use. With 2 options, uses might differ over many trials.
|
|
seen = set()
|
|
for i in range(50):
|
|
result = evaluate(parse("${x={a|b}} ${x} ${x}"), EvaluationContext(rng=random.Random(i), wildcard_manager=WildcardManager()))
|
|
words = [p for p in result.split() if p]
|
|
seen.add(tuple(words))
|
|
# Eventually we should see differing values
|
|
assert any(w[0] != w[1] for w in seen)
|
|
|
|
|
|
def test_variable_default(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("${undefined:fallback}"), ctx)
|
|
assert result == "fallback"
|
|
|
|
|
|
def test_nested_variant(ctx: EvaluationContext) -> None:
|
|
result = evaluate(parse("{a|{b|c}}"), ctx)
|
|
assert result in {"a", "b", "c"}
|
|
|
|
|
|
def test_cyclical_inline(ctx: EvaluationContext) -> None:
|
|
# {@ } in random evaluator uses ctx.cycle_counters — cycles within one prompt
|
|
r1 = evaluate(parse("{@x|y|z}"), ctx)
|
|
assert r1 in {"x", "y", "z"}
|