[CASSANDRA-21349] Fix backslash doubling in COPY TO/FROM round-trip on Python 3.10+

[howiezhao]

Jira Link: https://issues.apache.org/jira/browse/CASSANDRA-21349

Issue

COPY TO followed by COPY FROM corrupts string values containing backslashes on Python 3.10+: the backslash count doubles on every round-trip. For example, https:\/\/google.com becomes https:\\/\\/google.com after one cycle.

RC

There are two independent backslash-escaping layers in the COPY pipeline:

1. format_value_text (pre-doubling, always present): doubles every \ → \\ before passing to csv.writer
2. csv.writer escapechar handling (Python-version-dependent): Python 3.10 fixed bpo-12178, making csv.writer properly escape \ → \\ in all fields

Before Python 3.10, csv.writer did not escape bare backslashes in unquoted fields, so the pre-doubling from format_value_text was exactly cancelled by csv.reader on import. After Python 3.10, both layers apply, but csv.reader only removes one — leaving the value with doubled backslashes.

Fix

In ExportProcess.format_value, undo format_value_text's pre-doubling on Python 3.10+ before handing the value to csv.writer, making csv.writer the sole escaping layer:

if sys.version_info >= (3, 10):
    formatted = formatted.replace('\\\\', '\\')

No change for Python < 3.10.

[omniCoder77]

Do you have a jira for this?

[omniCoder77]

@howiezhao can you write test cases? refer to test_copyutil.py

[howiezhao]

Hi @omniCoder77

Do you have a jira for this?

Nope, should I create one?

@howiezhao can you write test cases? refer to test_copyutil.py

Thanks for the reminder, I've already added it.

[omniCoder77]

@howiezhao
https://chatgpt.com/share/69f79c28-cc80-8321-8067-9fdc3a752a26

[omniCoder77]

@howiezhao after ticket creation, update the PR description to include the link to jira ticket