export() raises UnicodeEncodeError and interpolate_file() returns a corrupted prompt when the locale is not UTF-8

[ebarkhordar]

On a machine whose locale is not UTF-8 (LC_ALL=en_US.ISO-8859-1 here), two things break as soon as the text has non-ASCII in it.

chat.export() raises instead of writing the transcript, for both .md and .html.

interpolate_file() is the quieter one. It returns a corrupted prompt rather than raising, so a UTF-8 prompt file reaches the model as mojibake and nothing says so.

Repro, no API key needed:

import os, tempfile
from chatlas import ChatOpenAI, UserTurn, AssistantTurn, interpolate_file

TEXT = "Wie heißt die Hauptstadt? 中文测试 café"

chat = ChatOpenAI(api_key="not-used", model="gpt-4o")
chat.set_turns([UserTurn(TEXT), AssistantTurn(TEXT)])
try:
    chat.export(os.path.join(tempfile.mkdtemp(), "transcript.md"))
    print("export: ok")
except Exception as e:
    print("export:", type(e).__name__ + ":", e)

p = os.path.join(tempfile.mkdtemp(), "prompt.md")
with open(p, "w", encoding="utf-8") as f:
    f.write(TEXT + " {{ name }}")
print("interpolate_file gives back what the file says:",
      TEXT in interpolate_file(p, variables={"name": "x"}))

Under LC_ALL=en_US.ISO-8859-1:

export: UnicodeEncodeError: 'latin-1' codec can't encode characters in position 35-38: ordinal not in range(256)
interpolate_file gives back what the file says: False

Same script under LC_ALL=en_US.UTF-8:

export: ok
interpolate_file gives back what the file says: True

One note on the second check: do not print the interpolated string to compare it by eye. Under a latin-1 stdout it re-encodes back to the original bytes and looks correct in the terminal, which is how I missed it the first time.

python:3.12-slim, Python 3.12.14, chatlas 0.23.0 from PyPI and 7b3aa56 on main, identical on both.

I only hit this in a container. The case that probably matters is Windows, where cp1252 is the default and a prompt file with an accent or a curly quote in it is completely ordinary. I have not tried it there.