how would you go about capturing a block comment? about funcj HOT 4 CLOSED

Hi,

I suspect the problem is that if your parser operates directly on the chars (as opposed to a token stream), then this grammar is not LL(1). When the parser encouters a "*" within the comment, it needs to look an extra char ahead to verify that "/" is the following char and therefore we are at the comment end. This makes it an LL(2) grammar, which this parser doesn't support.

One workardound is to first convert the char stream to a token stream, such that "/" and "/" become single tokens (e.g. COMMENT_START and COMMENT_END).

Some parsers get around this restriction by allowing backtracking. The problem with this is when the parser encounters an actual syntax error it then backtracks to see if there exists another route, meaning you then lose the location at which the error occurred, and the resultant error message is usually meaningless.

I'll have a play around with this to see whether there are any other solutions.

from funcj.

I got a hacky solution working but it has flaws:

public static final Parser<Chr, String> blockComment() {
        final String[] prev = {""};
        return string("/*")
                .andR(satisfy("closing block comment", c -> {
                    if (c.equals('/') && prev[0] == "*") {
                        return false;
                    } else {
                        prev[0] = String.valueOf(c);
                        return true;
                    }
                }).many())
                .map(Chr:: listToString)
                .map(s -> s.substring(0, s.length() - 2));
    }

The main issue is that you have to reconstruct the parser every time it is used.

from funcj.

Does look somewhat hacky, but if it works...

I'll close this issue for now.

from funcj.

It is totally hacky, I am not proud of it :)

To your original comment, I think I agree ... the performance hit might be minimal and the greater flexibility might be worth it.

from funcj.