af36f409b0
Co-authored-by: Dr. Carlos Daniel Carvalho <292283903+Dandandad@users.noreply.github.com> Co-authored-by: nesquena-hermes <nesquena+hermes@gmail.com>
124 lines
3.6 KiB
Python
124 lines
3.6 KiB
Python
"""Regression coverage for large-session fuzzy duplicate matching."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
|
|
def _visible_keys(count: int) -> set[tuple]:
|
|
return {("assistant", f"message {idx}", "") for idx in range(count)}
|
|
|
|
|
|
def test_large_visible_set_keeps_exact_duplicate_matching():
|
|
import api.models as models
|
|
|
|
keys = _visible_keys(1001)
|
|
exact = ("assistant", "message 1000", "")
|
|
|
|
assert models._matching_visible_duplicate(exact, keys) == exact
|
|
|
|
|
|
def test_large_visible_set_skips_fuzzy_duplicate_scan(monkeypatch):
|
|
import api.models as models
|
|
|
|
keys = _visible_keys(1001)
|
|
fuzzy_only = ("assistant", "message 1000 with suffix", "")
|
|
|
|
def unexpected_lookup(_keys):
|
|
raise AssertionError("large visible sets must not build the fuzzy lookup")
|
|
|
|
monkeypatch.setattr(models, "_build_visible_duplicate_lookup", unexpected_lookup)
|
|
|
|
assert models._matching_visible_duplicate(fuzzy_only, keys) is None
|
|
|
|
|
|
def test_threshold_sized_visible_set_preserves_fuzzy_recovery():
|
|
import api.models as models
|
|
|
|
keys = _visible_keys(1000)
|
|
fuzzy_only = ("assistant", "message 999 with suffix", "")
|
|
|
|
assert models._matching_visible_duplicate(fuzzy_only, keys) in keys
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
"workspace_prefix",
|
|
("[Workspace::v1: /tmp/project]", "[Workspace: /tmp/project]"),
|
|
)
|
|
def test_large_merge_dedupes_workspace_prefixed_middle_replay(workspace_prefix):
|
|
import api.models as models
|
|
|
|
sidecar = [
|
|
{
|
|
"role": "user" if idx % 2 == 0 else "assistant",
|
|
"content": f"message {idx}",
|
|
"timestamp": float(idx),
|
|
}
|
|
for idx in range(1000)
|
|
]
|
|
sidecar.insert(
|
|
500,
|
|
{"role": "user", "content": "canonical prompt", "timestamp": 499.5},
|
|
)
|
|
sidecar.append(
|
|
{"role": "assistant", "content": "final assistant", "timestamp": 1001.0}
|
|
)
|
|
state_replay = [
|
|
{
|
|
"role": "user",
|
|
"content": f"{workspace_prefix}\ncanonical prompt",
|
|
"timestamp": 2000.0,
|
|
}
|
|
]
|
|
|
|
merged = models.merge_session_messages_append_only(sidecar, state_replay)
|
|
|
|
assert len(merged) == len(sidecar)
|
|
assert merged[-1]["role"] == "assistant"
|
|
assert merged[-1]["content"] == "final assistant"
|
|
|
|
|
|
def test_large_merge_preserves_literal_workspace_prefix_as_distinct_turn():
|
|
import api.models as models
|
|
|
|
trailing_text = "same trailing text"
|
|
literal_sidecar_text = f"[Workspace: user typed this literally]\n{trailing_text}"
|
|
sidecar = [
|
|
{
|
|
"role": "user" if idx % 2 == 0 else "assistant",
|
|
"content": f"message {idx}",
|
|
"timestamp": float(idx),
|
|
}
|
|
for idx in range(1000)
|
|
]
|
|
sidecar.insert(
|
|
500,
|
|
{"role": "user", "content": literal_sidecar_text, "timestamp": 499.5},
|
|
)
|
|
sidecar.append(
|
|
{"role": "assistant", "content": "stale assistant tail", "timestamp": 1001.0}
|
|
)
|
|
distinct_state_turn = {
|
|
"role": "user",
|
|
"content": f"[Workspace::v1: /tmp/actual]\n{trailing_text}",
|
|
"timestamp": 2000.0,
|
|
}
|
|
|
|
merged = models.merge_session_messages_append_only(
|
|
sidecar,
|
|
[distinct_state_turn],
|
|
)
|
|
|
|
assert len(merged) == len(sidecar) + 1
|
|
assert any(message.get("content") == literal_sidecar_text for message in merged)
|
|
assert merged[-1] is distinct_state_turn
|
|
|
|
|
|
def test_small_visible_set_preserves_fuzzy_recovery():
|
|
import api.models as models
|
|
|
|
keys = {("assistant", "canonical answer", "")}
|
|
fuzzy_only = ("assistant", "canonical answer with suffix", "")
|
|
|
|
assert models._matching_visible_duplicate(fuzzy_only, keys) in keys
|