Re: gnubol: New bison Grammar available (long)

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

>>>>> "Bob" == RKRayhawk  <RKRayhawk@aol.com>
>>>>> wrote the following on Mon, 27 Dec 1999 01:19:27 EST

  Bob> Our thinking is different here.  By the way where is Michael
  Bob> when we need him?

He's not gone, although he's been a little distracted lately for
reasons of little interest to the list.

I generally favor Tim's plan to use an AST as the primary interface
to semantics, if we can summon the courage to contemplate what that
means for an 75,000 line program.  If we're convinced that we can get
away with that, it's far and away the nicest presentation for
semantics.  I think we'd better assume, however, that the parser will
not be able to create ideal subtrees in every case, so it should be a
requirement that the tree be fairly easily manipulable in semantics.
I plan to investigate the Sorcerer technology of pccts RSN.

I owe you a full report on the pccts experiment, which I will have to
defer just a little longer.  A few words may be in order, though.

As you may be aware, I have chosen to try to use "dlg", the lexer
technology that pccts provides, because it integrates nicely with the
rest.  There is an inherent risk in this course, since dlg is a
traditional deterministic automaton and has no provision for
backtracking or more than one character look-ahead.  I found this
sufficiently worrisome that I felt compelled to see if I could really
make the thing run, so that's what I've been doing lately.

With a little judicious hackery, I've been able to overcome the
trailing period problem in pictures and numeric literals, which was my
primary concern.  It ain't very pretty, but it seems to work ok.  The
fallback would be to interface Flex to pccts, which is still a
possibility and has been done before. 

With a somewhat working lexer in place, I've been able to push a few
small programs through the parser.  Error recovery is still
non-existent, but, after correcting a few oversights, I have been
able to get the expected results for correct programs and some of the
incorrect ones.

More later.

Best regards,

Mike






--
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.