Re: Stack Info or subparsing?

[Akim Demaille]

Le 11 oct. 2012 à 15:20, Adam Smalin a écrit :

> Lets say I want to do something like this
> 
> @Start Blah
> int a,b
> void func() {
>    beep()
> }
> float c
> @end
> 
> Then I will use Blah in place of declaring all that. Kind of like a C
> define. Now here is the twist
> 
> @Start Blah
> int a,b
> void func() {
>    beep()
> 
> float c
> @end
> 
> I want this to fail because it doesn't end in the same scope. Doing } }
> should fail as well as it end one scope to many.
> What is the best way to implement this? I believe I can use unput when I
> see Blah but the problem is ensuring the same scope. If I do it in multiple
> code blocks I believe ending to little or many would still have the parser
> in the same state so I can't simply look at state. I'd need a way to look
> at the stack AND state. An easy hack would be to count '{' and '}' however
> that will not work because what if i ended it with float c=.
> 
> So my question is how do I look at the state so I know it wasn't ended at
> something like '=' (theres not that many valid states to do this. I believe
> only in the body of namespace, function and class). How do I look at the
> stack to know if I am in the right depth?

You seem to hesitate between dealing with this at the scanner
level (hence your words "state", "stack", "unput" etc.), and doing it
at the level of the parser (in which case this is trivial,
just write your grammar, it will work).

If you just want to balance the braces between @start and @end,
then doing it with Flex seems the right things, and just count
the braces.  If the content of @start/@end is well defined, and
should be controlled, then do that with Bison, and trust it:
it won't accept improperly balanced input.