I have updated the branch to not raise an error. However I must unconditionally warn, as it is not appropriate to swallow such errors as we have been. Your users can fix the warning by updating AsciiDoctor. I know this is not ideal for your project, @mojavelinux, but we do have other users and I would like to prevent this situation in the future.

I can compromise on the warning. Since it's only going to happen for the PHP language (as the code in Asciidoctor stands), it's contained enough to be manageable until the new versions of Asciidoctor roll out.

@jneen Since there's a chance that Lexer#lex can be called multiple times with a truthy opts, using (opts.keys - [:continue]) is going to generate temporary arrays just to be checked if the result is empty.
Instead, the suggested route by @mojavelinux if opts.size == 1 && opts[:continue] has no such side-effects.

@ashmaroli I don't think it will matter in terms of actual usage given how rare someone calling Lexer#lex with an opts hash should be.

@jneen This looks fine to merge to me. Are you OK with me squashing and merging?

Can we merge this? The current released version is very broken. Note:

    def lex(string, opts={}, &b)
      unless opts.nil?
        warn 'The use of opts with Lexer.lex is deprecated' if $VERBOSE
      end

the opts argument is defaulted to {}, so in the happy case it will not be nil, and a warning will be issued.

I'm in support of it.

@ashmaroli while i appreciate your concern, I believe the urgency of warning at the appropriate times is a much greater concern than a possible slight performance degradation for users who have written custom lexers that use lex :continue => true and not the publicly documented delegate API.

(for all other users, it is only a single nil check)