Trailing-context is not working about jflex HOT 3 CLOSED

Can you post a spec file with input that doesn't work for you as expected? The rules above are not in the JFlex syntax.

I'd expect the following to work:

%%

%public
%class Test
%debug

letters = [a-zA-Z]+
end = "stop"

%%

{letters} / {end}  { return 0; }

[^] { return 1; }

For input

somethingstop

you should get a single match for something, returning 0, and single matches afterwards for s, t, o, p.

from jflex.

Sorry for being much too short.
Your example can be compiled.
Taking this as a starting point, I tried to use trailing-context in a macro, see the example below. This is not supported and results in the described syntax error.

%%

%public
%class Test
%debug

letters = [a-zA-Z]+
end = "stop"

stopped = {letters} / {end}

%%

stopped  { return 0; }

[^] { return 1; }

It seems that, as a special case, trailing context regular expressions cannot be used in macro definitions.

I think that it would make it simpler to define complex lexical specs if that would be no special case, so I propose that as a new feature for JFlex: Offer support for trailing context rules in macros.

Background: The lexer I'm working on can be seen here. This implements the lexical structure for the Scheme programming language v5. So far, it was possible to define all scanning expressions as macros, matching only the terminal tokens in the match section. This is now not longer the case.

The requirement for using trailing context came in with the new Scheme v7 feature of nested comments, I added the relevant part of the lexical structure spec (note the comment text rule):

<comment> −→ ; all subsequent characters up to a line ending
  | <nested comment>
  | #; <intertoken space> <datum>
<nested comment> −→ #| <comment text>
  <comment cont>* |#
<comment text> −→ <character sequence not containing
  #| or |#>
<comment cont> −→ <nested comment> <comment text>

from jflex.

It seems that, as a special case, trailing context regular expressions cannot be used in macro definitions.

That is indeed the case, and error reporting for it is not exactly great. I've opened #1089 for that.

I think that it would make it simpler to define complex lexical specs if that would be no special case, so I propose that as a new feature for JFlex: Offer support for trailing context rules in macros.

This is going to be tricky, and I'm not sure yet it's worth it. The underlying problem is that the matching algorithm only works for trailing context at the top level (otherwise we're not matching a regular language any more). If we allow trailing contexts in macros, we can now syntactically get trailing context at any nesting level in an expression.

I guess we could allow it syntactically and have a semantic pass for checking that it is only used at the top level.

from jflex.