Adding a feature `unicode` about lalrpop HOT 5 CLOSED

That sounds like a good idea.
Users of lalrpop don't need to know that it internally uses the regex crate, it's perfectly possible to only have a dependency on lalrpop.
Having to add a direct dependency on regex and regex-syntax just to enable some optional features on them is not intuitive. But a unicode feature on lalrpop is. It's also easy do document: If you use unicode character classes in your grammar, you need to add the unicode-feature on lalrpop.

from lalrpop.

Because lalrpop itself didn't require the feature, we couldn't catch the problem.

Small clarification, lalrpop does require this feature in CI (for example here, however because CI is ran with a Cargo.lock present, we never ran lalrpop CI with regex 1.8. Lalrpop binary users get the versions in lalrpop's Cargo.lock, but lalrpop library users do now.

Other than that, I think your explanation is spot on.

from lalrpop.

see also #750 (comment)

Thanks @jannic

from lalrpop.

This makes sense to me, though I would appreciate from somebody a summary of exactly what this breakage is all about.

from lalrpop.

@nikomatsakis by looking #750, user can input any regex. Here is my guess based on it.

Until regex 1.7, even if your input required unicode feature of regex-syntax, it passed because lalrpop implicitly had dependency to regex-syntax with unicode feature.
With regex 1.8, it is broken. regex 1.8 requires regex-syntax 0.7 while lalrpop depends on regex-syntax 0.6. Our dependency to regex(implicitly 1.8) enabled regex-syntax 0.7 with unicode feature, but the direct dependnecy regex-syntax 0.6 with no feature. Because lalrpop itself didn't require the feature, we couldn't catch the problem.
Users who depending on regex-syntax with unicode features started to experience problem because unicode feature is unexpectedly and silently dropped in their dependency.

from lalrpop.