Glob specification is not clear

[Conaclos]

This is related to editorconfig/specification#45.

When reading the Glob Expression section of the specification, I have several questions unanswered:

1. Are globs unicode-aware?
   Globs from .gitIgnore are not unicode-aware.
   This means that the globs [é] and ? don't match é.
2. How is handled three or more stars ***?
   Usually glob libraries treat it as a single star, or as an invalid glob expression.
3. Can character class match against /?
   Usually glob libraries don't allow that.
4. Can {s1,s2,s3} contain nested glob expressions? e.g. {*.json,*.toml}.
5. Can {s1,s2,s3} contain empty strings? e.g. {*.gen,}.json.
6. Does {num1..num2} allow reversed ranges such as {1..0}?
7. Does {num1..num2} allow numbers with leading zeros such as {01..02}
   If yes, is this equivalent to [12] or 0[12]?
8. Does {num1..num2} allow numbers with a + sign? e.g. {+1..2}?

Thanks!

[florianb]

@Conaclos thanks for bringing this to our attention. I think the specification is unclear at that point because I'm fact two separate technologies seem to apply: glob and brace expansion.

I need to do further research on that topic.

[Conaclos]

Feel free to tag me if you need a review or some help.

[cr7pt0gr4ph7]

The editorconfig/editorconfig-core-test library provides a set of tests that is used for the editorconfig-core-c, editorconfig-core-ruby, editorconfig-core-py etc. libraries.

The behavior specified through the glob and filetree tests, which also include comments, is probably the closest there currently is to an official specification. It should be possible to to turn these into an offical specification, or at least add a (normative or non-normative) reference to them in the spec.

Maybe @cxw42 knows more, as they were the last to modify those tests.

[cxw42]

Are globs unicode-aware?

I don't think we test for unicode, so it depends on the regex engine used by your core. I agree it would be good to specify this and add tests.

How is handled three or more stars ***?

Another we don't test for, so it's core-specific. In the C core (reference implementation), it's treated like ** followed by *.

Can character class match against /?

Yes, and we do test for this. [/] matches that literal sequence of three characters. It appears that is not adequately documented in the spec.

Can {s1,s2,s3} contain nested glob expressions? e.g. {.json,.toml}.

Yes

Can {s1,s2,s3} contain empty strings? e.g. {*.gen,}.json.

Yes

Does {num1..num2} allow reversed ranges such as {1..0}?

Not in the C core, but I don't see that we test for that.

Does {num1..num2} allow numbers with leading zeros such as {01..02}

Not in the C core, but I don't see that we test for that.

Does {num1..num2} allow numbers with a + sign? e.g. {+1..2}?

Yes, in the C core, but I don't see that we test for that.

Re. numeric ranges, see #371 for more discussion.

[cxw42]

Moving to the main issue tracker for visibility.