Multi-char escapes wrongly forbidden in character class

[Paul J. Lucas]

If you have a character range, e.g., A-Z, then the end-point chars in the range can be SingleCharEsc. A while ago, a "fix" was made for this, but the "fix" went too far and forbids MultiCharEsc within charClassExpr.

[Paul J. Lucas (paul-lucas)]

Removing the "fix" code results in the regex_err16.xq test failing. That test is:

  fn:matches("a", "[\s-e]")

The charClassExpr is invalid because, in character ranges, only SingleCharEsc are allowed and \s is a MultiCharEsc. ICU doesn't detect this and the test just returns "false."

Adding a proper fix for this would involve adding more state to the regex parser and knowing when we're within a character class *and* within a character range, i.e.:

  if ( in_char_class && c == '-' && prev_c_was_an_esc && !prev_c_was_a_single_char_esc )
    throw an exception

[Paul J. Lucas (paul-lucas)]

It actually harder than that, especially for detected the opposite case, e.g., "[e-\s]".

[Paul J. Lucas (paul-lucas)]

It's been decided to remove the "fix" and mark regex_err16.xq as an expected failure. A new bug #1023168 has been created.