fix: handle errors when decoding bytes from subprocesses (#222)

Since these bytes come from the output of unknown processes, there is no
guarantee that they are valid utf-8 text. We also have no way to know
what encoding they have (if any), so the best we can do is replace the
invalid bytes with the usual unicode marker for that.

Fixes #221

craft_cli/messages.py

@@ -239,8 +239,9 @@ def _write(self, data: bytes) -> None:
             useful_line = data[pointer:newline_position]
             pointer = newline_position + 1
 
-            # write the useful line to intended outputs
-            unicode_line = useful_line.decode("utf8")
+            # write the useful line to intended outputs. Decode with errors="replace"
+            # here because we don't know where this line is coming from.
+            unicode_line = useful_line.decode("utf8", errors="replace")
             # replace tabs with a set number of spaces so that the printer
             # can correctly count the characters.
             unicode_line = unicode_line.replace("\t", "  ")

tests/unit/test_messages_stream_cm.py

@@ -257,6 +257,24 @@ def test_pipereader_tabs(recording_printer, stream):
     assert msg.text == "::   123  456"  # tabs expanded into 2 spaces
 
 
+@pytest.mark.parametrize(
+    ("invalid_text", "expected"),
+    [(b"\xf0\x28\x8c\xbc", "�(��"), (b"\xf0\x90\x28\xbc", "�(�"), (b"\xf0\x90\x8c\x28", "�(")],
+)
+def test_pipereader_invalid_utf8(recording_printer, invalid_text, expected):
+    """Check that bytes that aren't valid utf-8 text don't crash."""
+    invalid_bytes = b"valid prefix " + invalid_text + b" valid suffix\n"
+
+    flags = {"use_timestamp": False, "ephemeral": False, "end_line": True}
+    prt = _PipeReaderThread(recording_printer, sys.stdout, flags)
+    prt.start()
+    os.write(prt.write_pipe, invalid_bytes)
+    prt.stop()
+
+    (msg,) = recording_printer.written_terminal_lines
+    assert msg.text == f":: valid prefix {expected} valid suffix"
+
+
 def test_pipereader_chunk_assembler(recording_printer, monkeypatch):
     """Converts ok arbitrary chunks to lines."""
     monkeypatch.setattr(messages, "_PIPE_READER_CHUNK_SIZE", 5)