Hi again,
I want to share my solution with you and the future generations
stumbling across similar problems ;) I decided implementing the exact
semicolon rules was not worth the trouble:
- The grammar is already quite complex and difficult to understand,
changing these to respect semicolon rules is a waste of effort
- Let bison instruct the lexer when to inject a semicolon, like
<<OptSemicolon: { driver->inject_semicolon = 1; } ';'>>, did not work
out, because when the mid-action is executed, bison already has read the
next token, so it is to late to inject. But moving this action one token
earlier would cause the lexer to inject semicolon in the wrong places.
So, I just made all semicolons optional and tried to cope with the
conflicts. I just had two kinds of conflicts ( \o/ ), actually. One
--probably a quite common-- was:
Stmt:
...
| "return"
| "return" Expression
| Assignment
....
Expression and Assignment both may begin with an identifier. So bison
does not know whether to shift for the second option or to reduce the
first and start with the Assignment. The default --shifting the
Expression-- totally works for me. First because, having an Assignment
right after a `return' does virtually never appear, secondly the
type-checker is able dectect a `mis-shifted' return and could suggest an
explicit semicolon.
Problem solved. Thank you a lot for your support :)
Cheers,
Matthias
_______________________________________________
help-bison@gnu.org https://lists.gnu.org/mailman/listinfo/help-bison
