Re: A generic reader for Guile?

[Chris Bitmead <chris@tech.com.au>]

Let me get this straight...

You are saying that you could link the bison/lex skeleton into guile
(once and for all), and then define the rules for the language in Scheme
to generate a parser which is just as fast as a regular bison/lex
parser?

Then you are further saying that you could define the Scheme language
itself in Scheme generating bison/lex rules and therefore discard the
existing Scheme parser?

It all sounds incredibly clever. I might even have time to work on it
say 10 or 15 summers hence :-)


Mikael Djurfeldt wrote:
> 
> I had planned to surprise you guys with a Christmas present this year.
> But I was so lazy during Christmas so that I only got half-finished.
> Now I realize that it will be too long time before I can finish it, so
> I'll post the idea in case someone else feels like working on it.
> 
> Here's what I did:
> 
> I ran flex for a simple specification file.  From this output I
> deduced what was fixed code from a template and what was variable
> (dependent upon the specification file).  It turned out that
> everything that was variable consisted either of C preprocessor
> macro definitions or tables of numbers.
> 
> So I thought: Why not turn the macros into variables and write Scheme
> level accessors for them, and turn the tables into uniform vectors?
> Then it is possible to install the tables from the Scheme level.
> 
> Look what we have!  Ta ta!
> 
> An extremely efficient, dynamically configurable generic lexer!
> 
> In order to generate proper numbers for the tables I wrote a Scheme
> program which converted a Scheme spec into a suitable flex input file,
> ran flex and extracted the numbers from the flex output.
> 
> It worked!
> 
> Then I did the same for bison, but I never had time to finish this
> stage.
> 
> I had planned to write the rules corresponding to Guile's syntax and
> then surprise you all.  I think that the resulting parser would be
> faster than Guile's own.  8-)
> 
> Now, *this* was a hack, but I think the idea of a dynamically
> configurable table-based parser is great: you write the time critical
> inner loops in C and crunch the rules from the grammar into tables.
> 
> Wouldn't this be great as a basis for the multi-language support in
> Guile?
> 
> Wouldn't YOU, dear reader, find it just great to spend your summer
> holiday reading the table-based parser section in Aho's dragon book,
> studying the tricks used in flex and bison to achieve optimum
> performance and implement this idea?  :-)
> 
> It may even be possible to port the template code directly from
> flex/bison and port the C code for generating the tables to Scheme.
> 
> Maybe we should divide it into two projects?  One scanner (flex) and
> one parser (bison).