Follow up on an idea in #31: use the Racket lexer to lex double-quoted strings in a grammar, thereby giving them all the semantics of Racket strings, and preventing any corner cases. This solution can’t be extended to single-quoted strings, however, because even with readtable tomfoolery to recognize a single quote as an opening delimiter for a string, the string still needs to be closed with a double quote. (I don’t know why this should be so, but it is the documented behavior.) Therefore single-quoted strings still are subject to the homegrown lexing solution and the flaws therein. Still, I don’t see that single-quoted strings have ever been a documented feature of brag (or its predecessor ragg). Perhaps they could be dropped altogether. For now, this solution is satisfying, because double-quoted strings are the dominant notation, and this PR will make them as good as they can be.
This improves the lexing of escape sequences within strings that appear in a grammar. It relies on Racket’s `read` to interpret these escape sequences rather than a hard-coded hash table. This gives strings in a grammar pretty much the same semantics as standard Racket strings, including support for octal and hex escape sequences for Unicode codepoints.
Though this passes all current tests, there are still some oddball corner cases that can be discovered by sticking together certain combinations of escape sequences (backslashes, double quotes, and codepoints). The better solution would be to peek into the input port for a double quote, and if it’s there, use the standard Racket lexer to pull out the string (this lexer already handles the weirdo cases). We can’t do this, however, because brag also supports single-quoted strings, which need to have the same semantics, and the Racket lexer won’t work with those. So I think we’re stuck with the homegrown solution (for consistency with both kinds of quotes) even at the expense of a few unresolved corner cases. Let’s leave that question for another day, as these cases haven’t surfaced in practical use thus far.
When using `make-rule-parser` for any rule that is not the start
rule, and applying the resulting parser while giving it a source-path
it would just use the entire grammar from the start rule.
Matthew Flatt
Matthew Butterick