quoted-strings only-when-needed incorrectly detecting strings with redundant quotes

[RealOrangeOne]

Since upgrading to 1.21.0, I enabled the following config:

  quoted-strings:
    quote-type: double
    required: only-when-needed

Unfortunately, seems there's some strings which it believes don't need quoting, when in fact they do:

• aur_builder ALL=(ALL) NOPASSWD: /usr/bin/pacman
• %H:%M:%S
• 'Mac' in ansible_facts.product_name

It seems to be something around when a string either contains or starts with a special character, but isn't entirely a special character?

[adrienverge]

Hi @RealOrangeOne (cc @ruipinge), thanks for the report and sorry about that.

There are actually 2 problems:

1. Strings that actually need quotes, but only-when-needed reports them as "redundantly quoted":
   - "aur_builder ALL=(ALL) NOPASSWD: /usr/bin/pacman"
   - "%H:%M:%S"
   - "'Mac' in ansible_facts.product_name"
2. Strings that don't need quotes, but undetected by only-when-needed:
   - "-----"
   - ":char"

I believe using

yamllint/yamllint/rules/quoted_strings.py

Lines 80 to 81 in 1a13837

	START_TOKENS = {'#', '*', '!', '?', '@', '`', '&',
	',', '-', '{', '}', '[', ']', ':'}

and checking token.value[0] in START_TOKENS is not a good way to do, because we can't ensure the need for quotes using only the first character.

Could an alternative solution be to drop START_TOKENS, and try parsing the value to check whether it would be seen as a string or something else?
Do you have any other idea?

[RealOrangeOne]

Perhaps, although there are complexities. For example, given foo: 6, should it be quoted or not?

Perhaps attempting to parse the value, and if it results in a complex type (ie list, object, parse error, whatever), then raise?

I'd be interested in knowing if there's a solution for the simpler data types (int, bool etc), but I can't think of one.

[adrienverge]

From a user perspective, foo: 6 isn't a problem: either it's quoted (therefore it's a string), or it's not (so it's a mapping). In either case, yamllint shouldn't complain.

I'm not sure I understand what you mean by "then raise".

To be clearer about my idea on required: only-when-needed: if a quoted string is found (e.g. "foo: bar" or "-----"), try to parse its contents as YAML.
→ If it results in a string (e.g. "-----" → "-----"), then warn "redundantly quoted".
→ If it results in another type (e.g. "foo: bar" → {foo: bar}), don't report anything.