Symptom ------- LLM requests intermittently fail with: 'utf-8' codec can't encode characters in position N-N+1: surrogates not allowed when messages contain emoji-heavy content (e.g. HTML with mixed emoji + JSON round-trips). This blocks the affected session until the session file is quarantined. Root cause ---------- Surrogate sanitization was only applied at the CLI entry point (nanobot/cli/commands.py: _sanitize_surrogates). Requests entering the LLM provider layer through other channels (Feishu, cron, webui, tool results, memory injection) had no defensive cleaning, so any message that happened to carry unpaired UTF-16 surrogates (from an upstream JSON round-trip with ensure_ascii=True on ill-formed input, memory rehydration, or third-party content) would blow up at json.dumps -> HTTP encode time inside the provider client. Fix --- 1. Extract sanitize_surrogates() and sanitize_surrogates_deep() into nanobot/utils/helpers.py as the single source of truth. Both use utf-16-le round-tripping with errors='surrogatepass' / 'replace', so paired surrogates reconstruct back into their real code point and lone surrogates collapse to U+FFFD. 2. Make nanobot/cli/commands.py:_sanitize_surrogates a thin wrapper that re-exports the shared helper (backward compatible). 3. Add defense-in-depth at the LLM provider boundary in nanobot/providers/base.py:_sanitize_empty_content by running sanitize_surrogates_deep over each message and its content blocks right before requests are serialized to JSON. Non-goals --------- - truncate_text() is intentionally left untouched. Python str slicing cannot split a single code point into surrogate halves, so it is not the source of lone surrogates. - session/manager storage layer is untouched. Archived sessions reproduced the failure only through the request path, not through storage. Verification ------------ - New regression suite tests/providers/test_sanitize_surrogates.py covers: paired surrogate reconstruction, lone surrogate replacement, identity return on clean input (zero allocation), deep recursion on dict/list/tuple, provider _sanitize_empty_content integration, and full utf-8 encodability of the sanitized request body. - 14/14 new tests pass; full existing test module also green. - Replayed 58 archived real session messages plus adversarial lone-surrogate injection through the provider path with no encode errors after the fix. Impact ------ - No behaviour change for clean inputs (sanitize_surrogates_deep is an identity return when no surrogate is present). - Fails-safe: unpaired surrogates degrade to U+FFFD instead of aborting the entire request.
175 lines
6.4 KiB
Python
175 lines
6.4 KiB
Python
"""Regression tests for lone UTF-16 surrogate scrubbing in provider payloads.
|
|
|
|
These lock down two behaviors:
|
|
|
|
1. ``nanobot.utils.helpers.sanitize_surrogates`` / ``sanitize_surrogates_deep``
|
|
produce strings that can round-trip through ``str.encode('utf-8')`` even
|
|
when the input contains unpaired surrogates.
|
|
|
|
2. ``LLMProvider._sanitize_empty_content`` applies the deep sanitize as a
|
|
defense-in-depth pass so that ``UnicodeEncodeError: 'utf-8' codec can't
|
|
encode characters ... surrogates not allowed`` cannot escape into the
|
|
HTTP client when messages contain emoji-heavy history plus a lone
|
|
surrogate leak (e.g. from Windows console input or a truncated JSON
|
|
round-trip).
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
from nanobot.providers.base import LLMProvider
|
|
from nanobot.utils.helpers import (
|
|
sanitize_surrogates,
|
|
sanitize_surrogates_deep,
|
|
)
|
|
|
|
|
|
class TestSanitizeSurrogates:
|
|
def test_paired_surrogates_reconstructed_to_emoji(self):
|
|
# \uD83E\uDD16 is the UTF-16 surrogate pair for 🤖 (U+1F916).
|
|
raw = "hello \ud83e\udd16 world"
|
|
cleaned = sanitize_surrogates(raw)
|
|
assert cleaned == "hello 🤖 world"
|
|
cleaned.encode("utf-8") # must not raise
|
|
|
|
def test_lone_surrogate_replaced_with_fffd(self):
|
|
raw = "hello \ud83e world" # lone high surrogate
|
|
cleaned = sanitize_surrogates(raw)
|
|
assert "\ud83e" not in cleaned
|
|
# replacement char U+FFFD substitutes the unpaired surrogate
|
|
assert "\ufffd" in cleaned
|
|
cleaned.encode("utf-8") # must not raise
|
|
|
|
def test_normal_text_returned_unchanged_identity(self):
|
|
raw = "plain ascii + 中文 + 🤖"
|
|
assert sanitize_surrogates(raw) == raw
|
|
|
|
def test_non_string_returned_as_is(self):
|
|
assert sanitize_surrogates(None) is None # type: ignore[arg-type]
|
|
assert sanitize_surrogates(42) == 42 # type: ignore[arg-type]
|
|
|
|
|
|
class TestSanitizeSurrogatesDeep:
|
|
def test_clean_input_returns_same_object(self):
|
|
payload = {"role": "user", "content": [{"type": "text", "text": "hi 🤖"}]}
|
|
result = sanitize_surrogates_deep(payload)
|
|
# No allocation on clean input: same object identity.
|
|
assert result is payload
|
|
|
|
def test_lone_surrogate_in_nested_content_cleaned(self):
|
|
dirty = {
|
|
"role": "user",
|
|
"content": [
|
|
{"type": "text", "text": "hello \ud83e world"},
|
|
{"type": "text", "text": "normal"},
|
|
],
|
|
}
|
|
cleaned = sanitize_surrogates_deep(dirty)
|
|
assert cleaned is not dirty # rebuilt
|
|
first = cleaned["content"][0]["text"]
|
|
assert "\ud83e" not in first
|
|
# Full payload must be UTF-8 encodable.
|
|
import json
|
|
|
|
json.dumps(cleaned).encode("utf-8")
|
|
|
|
def test_paired_surrogates_in_list_collapse_to_emoji(self):
|
|
dirty = ["a", "b \ud83e\udd16", "c"]
|
|
cleaned = sanitize_surrogates_deep(dirty)
|
|
assert cleaned == ["a", "b 🤖", "c"]
|
|
|
|
def test_deep_recursion_on_tuple_and_dict(self):
|
|
dirty = (
|
|
{"k": "\ud83e"},
|
|
["nested", "\ud83e\udd16"],
|
|
"\ud83e clean",
|
|
)
|
|
cleaned = sanitize_surrogates_deep(dirty)
|
|
# tuple preserved as tuple
|
|
assert isinstance(cleaned, tuple)
|
|
assert "\ud83e" not in cleaned[0]["k"]
|
|
assert cleaned[1][1] == "🤖"
|
|
|
|
|
|
class TestProviderSanitizeEmptyContent:
|
|
"""The provider-boundary defense-in-depth pass.
|
|
|
|
Lone surrogates leaking into any string leaf of a request message must
|
|
not survive ``_sanitize_empty_content``; otherwise the HTTP client will
|
|
raise ``UnicodeEncodeError`` when serializing the request body.
|
|
"""
|
|
|
|
def test_lone_surrogate_in_string_content_scrubbed(self):
|
|
messages = [
|
|
{"role": "user", "content": "leak \ud83e here"},
|
|
]
|
|
result = LLMProvider._sanitize_empty_content(messages)
|
|
text = result[0]["content"]
|
|
assert "\ud83e" not in text
|
|
text.encode("utf-8") # would raise pre-fix
|
|
|
|
def test_lone_surrogate_in_content_block_scrubbed(self):
|
|
messages = [
|
|
{
|
|
"role": "assistant",
|
|
"content": [
|
|
{"type": "text", "text": "hi \ud83e there"},
|
|
],
|
|
},
|
|
]
|
|
result = LLMProvider._sanitize_empty_content(messages)
|
|
text = result[0]["content"][0]["text"]
|
|
assert "\ud83e" not in text
|
|
text.encode("utf-8")
|
|
|
|
def test_paired_surrogates_reconstructed_in_content(self):
|
|
messages = [
|
|
{"role": "user", "content": "robot \ud83e\udd16 hello"},
|
|
]
|
|
result = LLMProvider._sanitize_empty_content(messages)
|
|
assert result[0]["content"] == "robot 🤖 hello"
|
|
|
|
def test_clean_input_semantics_unchanged(self):
|
|
messages = [
|
|
{"role": "user", "content": "plain 🤖 content"},
|
|
{"role": "assistant", "content": [{"type": "text", "text": "ok"}]},
|
|
]
|
|
result = LLMProvider._sanitize_empty_content(messages)
|
|
assert result[0]["content"] == "plain 🤖 content"
|
|
assert result[1]["content"][0]["text"] == "ok"
|
|
|
|
def test_full_request_body_is_utf8_encodable_after_sanitize(self):
|
|
"""End-to-end: an emoji-heavy history plus a lone surrogate must
|
|
no longer break utf-8 encoding of the outgoing HTTP request body."""
|
|
import json
|
|
|
|
messages = [
|
|
{"role": "system", "content": "You are a 🐱 assistant."},
|
|
{"role": "user", "content": "🤖 fixed?"},
|
|
{
|
|
"role": "assistant",
|
|
"content": [
|
|
{"type": "text", "text": "answer with a lone \ud83e half"},
|
|
],
|
|
},
|
|
]
|
|
cleaned = LLMProvider._sanitize_empty_content(messages)
|
|
# This is the operation that previously raised UnicodeEncodeError.
|
|
json.dumps({"model": "x", "messages": cleaned}).encode("utf-8")
|
|
|
|
|
|
class TestBackwardCompatReExport:
|
|
def test_cli_reexport_points_to_shared_helper(self):
|
|
"""The CLI module continues to expose ``_sanitize_surrogates`` as a
|
|
thin alias so existing imports (e.g. ``SafeFileHistory`` in the
|
|
legacy tests) keep working."""
|
|
from nanobot.cli.commands import _sanitize_surrogates as cli_alias
|
|
|
|
assert cli_alias is sanitize_surrogates
|
|
assert cli_alias("hello \ud83e\udd16") == "hello 🤖"
|
|
|
|
|
|
if __name__ == "__main__":
|
|
pytest.main([__file__, "-v"])
|