fix(hooks): keep a path's lone surrogates when a run log text is cut

src/chrys/service/hooks/run_log.py

@@ -107,14 +107,19 @@ def indented_json(value: object) -> str:
 def _clip_text(text: str, limit: int = HOOK_RUN_TEXT_MAX_BYTES) -> tuple[str, int]:
     """*text* cut to at most *limit* UTF-8 bytes, and its full size in them.
 
-    A cut drops the character it splits, and every lone surrogate of the text.
+    A cut drops the character it splits; lone surrogates, which a path that is not
+    UTF-8 holds, are kept as they are.
     """
     if text.isascii():
         return text[:limit], len(text)
     data = text.encode("utf-8", "surrogatepass")
     if len(data) <= limit:
         return text, len(data)
-    return data[:limit].decode("utf-8", "ignore"), len(data)
+    cut = limit
+    # Back to the first byte of the character the limit splits: its others are 10xxxxxx.
+    while cut and data[cut] & 0xC0 == 0x80:
+        cut -= 1
+    return data[:cut].decode("utf-8", "surrogatepass"), len(data)
 
 
 class HookRunLog:

tests/orchestration/engine/test_hook_run_reporting.py

@@ -24,7 +24,7 @@
 from chrys.service.hooks.run_log import HookRunLog
 from chrys.service.hooks.schema import HookConfig, HookExecution, HookMatch, HookRun, HookSettings, HooksFile
 from chrys.service.trajectory.hooks import HookOutcome
-from tests.support.waiting import ENGINE_TURN_TIMEOUT, wait_for, wait_until
+from tests.support.waiting import ENGINE_TURN_TIMEOUT, wait_for
 
 _SESSION = "s1"
 
@@ -267,7 +267,6 @@ async def explode(_event: HookRunsChanged) -> None:
     assert decision.blocked is True
     (record,) = log.records()
     assert (record.state, record.effect) == ("ended", "blocked")
-    assert not await wait_until(lambda: len(log.records()) != 1, timeout=0.2)
 
 
 def test_the_notice_does_not_reach_headless_or_acp_output() -> None:

tests/service/hooks/test_run_log_retention.py

@@ -541,10 +541,10 @@ def test_changed_arguments_past_the_cap_are_cut_with_their_size_kept(unit: str)
     detail = log.detail("rewrite")
     assert detail is not None and detail.applied is not None
     kept = detail.applied.args
-    assert len(kept.encode("utf-8", "surrogatepass")) <= HOOK_RUN_TEXT_MAX_BYTES
+    # Cut short of the limit by the character it would split, at most: a path that is not UTF-8 included.
+    assert HOOK_RUN_TEXT_MAX_BYTES - 4 < len(kept.encode("utf-8", "surrogatepass")) <= HOOK_RUN_TEXT_MAX_BYTES
     assert detail.applied.args_size == len(full.encode("utf-8", "surrogatepass"))
-    if unit != "\udc80":
-        assert full.startswith(kept)
+    assert full.startswith(kept)
 
 
 def test_changed_arguments_within_the_cap_are_kept_whole() -> None: