Style/RedundantRegexpEscape can generate incorrect fixes for hyphens in character classes

[bpo]

RedundantRegexpEscape is meant to detect and remove unneeded escapes for regular expressions. In certain cases an escaped hyphen in the middle of a character class can be automatically unescaped such that its meaning changes from a literal hyphen to creating a range match.

---

Expected behavior

A hyphen in the middle of a character class should only be unescaped if its resulting meaning stays the same.

Actual behavior

In at least some scenarios the hyphen can be converted to a range meaning. See below for a concrete example.

Steps to reproduce the problem

Example regexp: %r{[:\.\-/]*[^:\.\-/]+[:\.\-/]*}
This regexp could be safely reduced to %r{[:./-]*[^:./-]+[:./-]*} (moving the hyphen to the end of each class to avoid creating a range). Or one could leave the hyphen in place but leave it escaped (only un-escaping the period).
Test string "america:pnw.oregon-portland/kenton"
Using Ruby scan, results are: ["america:", "pnw.", "oregon-", "portland/", "kenton"]

Under this version of Rubocop the regexp above is incorrectly simplified to %r{[:\.-/]*[^:.\-/]+[:\.-/]*}
Ruby scan with this change instead yields: ["america:", "pnw.", "oregon", "portland/", "kenton"]

RuboCop version

$ bundle exec rubocop -V
0.88.0 (using Parser 2.7.1.4, rubocop-ast 0.1.0, running on ruby 2.6.6 x86_64-darwin19)

[marcandre]

Thanks for the report 👍

@owst do you feel like checking this out?

[owst]

Thanks for the heads-up @marcandre! PR opened that will fix this.

Apologies @bpo - this was caused by the leading : in the character classes tripping up some incorrect logic that was attempting to handle POSIX bracket expressions, which has now been simplified/corrected

[bpo]

👍 Nice @owst, thank you for the very handy feature and quick response