fix(release): decode distribution smoke output as UTF-8

scripts/verify_python_distribution.py

@@ -8,6 +8,7 @@
 import os
 import re
 import subprocess
+import sys
 import tarfile
 import tempfile
 import venv
@@ -22,6 +23,20 @@
 from packaging.version import InvalidVersion, Version
 
 REPOSITORY_ROOT = Path(__file__).resolve().parents[1]
+
+# ``python scripts/verify_python_distribution.py`` puts ``scripts/`` on
+# ``sys.path``, so the repository root is added explicitly before importing the
+# shared subprocess policy. Duplicating that policy here is how the verifier and
+# the runtime it checks would drift apart.
+if str(REPOSITORY_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPOSITORY_ROOT))
+
+from core.platform_compat import (
+    configure_utf8_stdio,
+    subprocess_env,
+    subprocess_text_kwargs,
+)
+
 PACKAGE_NAME = "deepcode-hku"
 MINIMUM_MCP_VERSION = Version("1.29")
 UNSUPPORTED_MCP_VERSION = Version("2")
@@ -294,11 +309,12 @@ def _run(
     stdin=None,
     timeout: int = 180,
 ) -> None:
-    environment = {
-        **os.environ,
-        "PIP_DISABLE_PIP_VERSION_CHECK": "1",
-        "PYTHONNOUSERSITE": "1",
-    }
+    environment = subprocess_env(
+        {
+            "PIP_DISABLE_PIP_VERSION_CHECK": "1",
+            "PYTHONNOUSERSITE": "1",
+        }
+    )
     try:
         subprocess.run(
             command,
@@ -307,7 +323,7 @@ def _run(
             stdin=stdin,
             check=True,
             capture_output=True,
-            text=True,
+            **subprocess_text_kwargs(),
             timeout=timeout,
         )
     except subprocess.CalledProcessError as exc:
@@ -381,6 +397,7 @@ def smoke_installed_wheel(wheel: Path) -> None:
 
 
 def main(argv: list[str] | None = None) -> int:
+    configure_utf8_stdio()
     parser = argparse.ArgumentParser(
         description="Verify DeepCode Python artifacts and their installed runtime."
     )

tests/test_python_distribution_release.py

@@ -4,6 +4,7 @@
 
 import importlib.util
 import json
+import subprocess
 import sys
 from pathlib import Path
 
@@ -111,3 +112,50 @@ def test_packaged_web_manifest_requires_its_entry_assets():
     del files["web/assets/app.js"]
     with pytest.raises(release.DistributionVerificationError, match="resource missing"):
         release.verify_web_assets(files.__getitem__, list(files), "web/", "2.2.0")
+
+
+def test_smoke_failure_reports_output_that_the_reader_cannot_decode(tmp_path):
+    """A rejected byte used to wipe out the whole diagnostic.
+
+    Reading without ``encoding`` decoded the child with the locale codec, so a
+    rejected byte raised inside subprocess's reader thread; the thread died, the
+    process result stayed clean, and ``_run`` reported the command with no
+    output at all. ``0x91`` is invalid UTF-8 and cp936, so the byte is rejected
+    on every runner.
+    """
+
+    code = (
+        "import sys; "
+        "sys.stdout.buffer.write(b'partial \\x91 output\\n'); "
+        "sys.stdout.buffer.flush(); "
+        "sys.exit(3)"
+    )
+    with pytest.raises(release.DistributionVerificationError) as failure:
+        release._run([sys.executable, "-c", code], cwd=tmp_path)
+
+    message = str(failure.value)
+    assert "partial" in message
+    assert "output" in message
+    assert "\ufffd" in message
+
+
+def test_smoke_children_are_told_to_write_utf8(monkeypatch, tmp_path):
+    """The read side declares UTF-8, so the child has to speak it too.
+
+    Otherwise CJK output from pip or the CLI degrades into replacement
+    characters that the release report cannot be read from.
+    """
+
+    observed = {}
+
+    def fake_run(command, **kwargs):
+        observed.update(kwargs)
+        raise subprocess.CalledProcessError(1, command, output="out", stderr="err")
+
+    monkeypatch.setattr(release.subprocess, "run", fake_run)
+    with pytest.raises(release.DistributionVerificationError, match="boom"):
+        release._run(["boom"], cwd=tmp_path)
+
+    assert observed["encoding"] == "utf-8"
+    assert observed["errors"] == "replace"
+    assert observed["env"]["PYTHONIOENCODING"] == "utf-8"