[breken-ai]

What type of PR is this? (check all applicable)

• Refactor
• Feature
• Bug Fix
• Optimization
• Documentation Update
• Other

Description

json_params_matcher decodes a bytes body as UTF-8. If that fails, it tries gzip.decompress(), which was added in #665. When a body is neither UTF-8 nor gzip, such as a binary file upload, that fallback raises BadGzipFile. A gzip body whose content isn't UTF-8 raises UnicodeDecodeError instead. Either way the exception comes out of the matcher and aborts the whole request, so other registered responses never get a chance to match:

url = "http://example.com/upload"
responses.post(url, match=[matchers.json_params_matcher({"name": "a.png"})])
responses.post(url, body="uploaded")

requests.post(url, data=b"\x89PNG\r\n\x1a\n\xff\x00")
# gzip.BadGzipFile: Not a gzipped file (b'\x89P')

This PR has the matcher catch those decode errors and return a mismatch with a reason, the same way it already handles JSONDecodeError. The request then falls through to the next response. If nothing matches, it gets the usual "Connection refused by Responses" message instead of a gzip traceback. Gzipped JSON bodies still match as before.

Related Issues

• None. I found no existing issue or PR for this. It is a follow-up to Support gzipped request.body in json_params_matcher #665.

PR checklist

Before submitting this pull request, I have done the following:

• Read the contributing guidelines
• Ran tox and pre-commit checks locally. I ran pre-commit on the changed files and pytest on Python 3.12 (urllib3 2) and Python 3.9 (urllib3 1.26): 253 passed. mypy shows the same 3 errors as master and none new.
• Added my changes to the CHANGES file

Added/updated tests?

Current repository has 100% test coverage.

• Yes. test_json_params_matcher_body_not_utf8_or_gzip fails on master and passes with this change. It covers a binary body falling through to a second response, and a gzip body that isn't UTF-8.
• No, and this is why:
• I need help with writing tests

Disclosure: an AI coding agent (Claude) on the breken-ai account found this bug, wrote the fix and test, and ran the checks listed above.