[rahul05ranjan]

Fixes #140.

Problem

dumps() documents quote_style as a QuoteStyle enum member, but every member's value is a plain string ('always_double', 'always_single', 'prefer_double', 'prefer_single'), so passing the value instead of the member is an easy mistake — and it is accepted silently. Because _encode_quoted_str() compares quote_style against the members with in/==, a raw string matches none of them:

• escaping decisions fall through to the "double-quote" branch while the delimiter check falls through to the single-quote branch, so quote_style='always_double' emits a single-quoted string with no quotes escaped;
• quote_style='always_single' emits a single-quoted string in which single quotes are not escaped;
• quote_style='prefer_double'/'prefer_single' silently never apply their preference (the flip/compare logic never matches).

The escaping guess and the delimiter can therefore contradict each other and produce output that json5.loads() itself rejects:

>>> json5.dumps("it's", quote_style='always_double')
"'it's'"
>>> json5.loads("'it's'")
ValueError: <string>:1 Unexpected "'" at column 5

Fix

In JSON5Encoder.__init__, coerce plain-string values into QuoteStyle members (QuoteStyle('always_double') → QuoteStyle.ALWAYS_DOUBLE) and raise a ValueError listing the valid values for strings that do not correspond to a member:

>>> json5.dumps("it's", quote_style='always_double')
'"it\'s"'
>>> json5.dumps("it's", quote_style='always-quote')
ValueError: 'always-quote' is not a valid QuoteStyle; expected one of: 'always_double', 'always_single', 'prefer_double', 'prefer_single'

This mirrors what json5/tool.py already does with its QUOTE_STYLES mapping, and leaves non-string custom values untouched for backwards compatibility.

Verification

• New regression test test_quote_style_string_is_coerced asserts, for all four members, that passing member.value produces byte-identical output to passing member (including a string containing both quote kinds), that both 'always_double' and 'always_single' now round-trip through loads(), and that an invalid string raises ValueError.
• Full local suite: python3 -m pytest tests/lib_test.py → 74 passed (73 pre-existing + 1 new).
• Noted in passing: tests/lib_test.py.new and tests/lib_test.py.new.2 look like accidentally committed editor leftovers on main — happy to remove them in a separate cleanup PR if wanted.

[dpranke]

As noted in a comment on the bug, I think I'd rather have this just fail rather accept the coercion. Does that make sense, and do you feel like changing things accordingly?

[rahul05ranjan]

That makes sense — done in acdbaf2. The coercion is gone: any value that is not a QuoteStyle member now raises ValueError listing the valid members, including a string that names or equals a member's value. Since the parameter is typed as QuoteStyle, non-string values fail the same way rather than silently falling back to double-quote behavior; happy to relax that if you'd rather keep the old leniency for non-strings.

[dpranke]

I don't think we need this comment; it's clear enough from the code what it's doing.

As for the error message, I appreciate the effort, but it's more detailed than I would like.
I'd probably use something like ValueError('quote_style argument must be a json5.QuoteStyle value'). We don't need to list each individual value

[dpranke]

I think these tests are more exhaustive than they need to be; just test that a json5.QuoteStyle value works and something that isn't a QuoteStyle doesn't. If you really want to be paranoid, you could have two failing tests, one for a 'always_double' and one for 1 or some other value.

You also don't need the tests on lines 1161-1168; working values are covered by other existing tests (specifically test_string_quote_styles.

So, I'd either move these tests inside that method, or create a new method that is immediately next to that method, and call it test_invalid_quote_styles; your name is more specific than it needs to be as well (since we might want to test for other non-strings).

[dpranke]

Thanks for the updates! I have some comments about the wording and testing, but otherwise things are close :).

[rahul05ranjan]

Thanks for the re-review! Addressed in d7f9f1c:

• Dropped the comment and simplified the error message to ValueError('quote_style argument must be a json5.QuoteStyle value') without listing the individual values.
• Replaced the rejection tests with a minimal test_invalid_quote_styles placed right next to test_string_quote_styles, covering the two failure cases you mentioned ('always_double' and 1); the working-value coverage is left to the existing tests.

Full suite passes locally (74/74).

[dpranke]

Looks good, thank you!

[rahul05ranjan]

Appreciate the merge on #141 — nice to see "fix: coerce string quote_style values into QuoteStyle members" ship.

On a personal note: I'm on the lookout for my next role (Full Stack (PHP, NodeJs, Python) / Data Science / Applied AI / Devtools, based in India). Happy to chat if there's an opening on your side or one you'd recommend:
LinkedIn: https://www.linkedin.com/in/rahulswt7/ · GitHub: https://github.com/rahul05ranjan

[dpranke]

On a personal note: I'm on the lookout for my next role

Sure :). It's been nice working with you. I've got nothing for you at the moment, but I'll try to keep you in mind if I hear of anything.

[rahul05ranjan]

Thanks Dirk, really appreciate it — and same here, this was a genuinely fun one to work on. Your reviews were quick, specific, and easy to act on, which made the whole loop a pleasure.

Zero pressure on the role front: if something crosses your radar I'd love a pointer, and if not, the collaboration was well worth it on its own. 🙂

All the best with pyjson5!