Re: gnubol: How are grammars progressing

[Tim Josling <tej@melbpc.org.au>]

Hi Bob,

The bison grammar for Nucleus level 1 ++ is done, but I have not
yet coded the kludges. United Airlines (there was actually a good
looking hostess on one flight- although she had the same attitide
problem endemic to UA) delivered me home safely, and the jetlag
is almost back to acceptable levels. I will probably start on the
kludges tomorrow or the day after. My plan is to have it done
before I go back to work on 5 Jan 2000.

The refmod kludge in the preprocessor was only about 100 lines,
so the overall task is not too bad. I have to build a subset of
the symbol to distinguish conditionals, class conditions and
variables within their hierarchies. SMOP - I hope.


Once that's done I'll post it, with a few test cases. Then hook
into the gcc back end and add things one at a time. I think it is
important to get something running that can then be added to
incrementally, now that the R&D is out of the way (will be once I
prove the kludges can be done).

The GCC interface might take a while but that will be after I
post the grammar. I think it will be more thinking than coding.

Life has a few more distractions back home so the rate of
progress is going to slow down. I have a few more options than 

a) Hotel room - TV
b) Hotel room - Coding
c) Various bars, blues clubs etc. 

I also have to finish painting the house (tomorrow!).

I've done a few cleanups and some checking and minor improvements
eg Bill's suggestion to allow perform if ... end-perform; bug
fixes to the preprocessor. 

I tried to find a reasonable way to allow evaluate (no end-verb)
and search (ditto) as nested conditionals but it is messy so I
dropped the idea for the moment.

I have been reading the list. Clearly as you suggested the parser
tools need some help. My approach is to change the tokens to
provide that help. You are suggesting that code in the
productions should run the overall show, and have the parser
tools only do the basic phrases. In the end, I don't think the
approaches differ much in practice. 

At the moment I feel confident that bison is the way to go.
Backtracking makes it hard to be sure what the parser will
actually do, and makes it impossible to provide good error
messages - one thing the existing compilers do. Also there is a
support issue with btyacc, inasmuchas he did not respond to my
patches to allow btyacc to generate C code (in spite of
suggestions that he would welcome them). I'll leave antlr to
others to test.

One thing others can do is look over the other verbs (outside the
nucleus) and point out any tricky things (eg invalid key is only
allowed on IO to certain types of files).

As usual no promises. My objective though is to have something
which is respectable by end 2000.

Tim Josling

> Have the bison or PCCTS grammars progressed?
> RKRayhawk@aol.com
>

--
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.