Re: gnubol: Re: Magic tokens

[Michael McKernan <mck@tivoli.mv.com>]

Tim, I'm going to play these straight.  While I think IBM might very
well have permitted some nested constructs, I'd be very surprised if
thay did this one.  They're lazier than you are.   Letting SIZE
ERROR nest is actually a little easier than handling it, but this one
takes a lot of effort and I don't imagine anyone wants it.  One case
that might fall out may be a WHEN after a WHEN OTHER, but I don't
want to do anything about it in general.  Let the syntax errors fall
where they may.

If it's a challenge, of course, I could probably get through it with
lookahead, but I don't really see any reason to do it.  My current
attitude probably makes my grammars for these verbs uninteresting, so
I won't inflict them on you.

What are you doing about illegal NEXT SENTENCE?  I find those
particularly nasty, and I've had to go outside of syntax proper to
catch them.

Regards,

Mike

>>>>> "Tim" == Tim Josling <TIMJOSLING@prodigy.net>
>>>>> wrote the following on Mon, 06 Dec 1999 08:45:50 +1000

  Tim> The problem arises because both search and inspect use WHEN as
  Tim> a delimiteer.  Imagine a simpler version of the grammar:

  Tim> %start statement

  Tim> statement: search_statement |evaluate_statement
  Tim> |display_statement ;

  Tim> search_statement: SEARCH (WHEN S statement)* ;

  Tim> evaluate_statement: EVALUATE (WHEN E statement)* 
  Tim> (WHEN OTHER statement)?  ;

  Tim> display_statement: DISPLAY literal ;

  Tim> So parse this:

SEARCH
     WHEN S
          EVALUATE
                   WHEN OTHER
     WHEN S
          EVALUATE
                    WHEN E
                        DISPLAY "hello hello hello"

  Tim> How do you tell who owns the WHENs with k=1 lookahead?

  Tim> In the real language, maybe with k>1 lookahead you can work it
  Tim> out, as suggested by the 'S' and 'E' which are meant to be
  Tim> grammar constructs that may be distinguishable or may not, but
  Tim> not with k=1.

  Tim> Once I have the parser running I can do some more practical
  Tim> experiments. To repeat though, it is not a problem unless IBM
  Tim> has support for the conditional forms of these verbs.




--
This message was sent through the gnu-cobol mailing list.  To remove yourself
from this mailing list, send a message to majordomo@lusars.net with the
words "unsubscribe gnu-cobol" in the message body.  For more information on
the GNU COBOL project, send mail to gnu-cobol-owner@lusars.net.