Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!decwrl!ucbvax!CS.ARIZONA.EDU!kwalker
From: kwalker@CS.ARIZONA.EDU (Kenneth Walker)
Newsgroups: comp.lang.icon
Subject: RE: fits posting; new l_scan() routines
Message-ID: <9008032305.AA28394@megaron.cs.arizona.edu>
Date: 3 Aug 90 23:05:32 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Distribution: inet
Organization: The Internet
Lines: 72

> Date: 3 Aug 90 16:19:17 GMT
> From: ux1.cso.uiuc.edu!midway!sophist!goer@iuvax.cs.indiana.edu

> I want very much to prototype this new scanning method using user-
> defined control structures, but it's proving very hard.

Emulating regular string scanning using user-defined control structures
is also awkward. There is a way to emulate string scanning using nested
procedure calls. It does not correctly handle breaking out of string
scanning (but then neither did the Icon intepreter until V7!).

The idea is to translate

   expr1 ? expr2

into

   escan(bscan(expr1), expr2)

(Of course, expr2 can be a compound expression, if you want it to be.)
This allows the setup routine, bscan, to comunicate with the termination
routine, escan. It allows nesting of scanning along with full backtracking
into and out of scanning. In addition, escan returns the value of the
scanning expression.

Saved scanning environments are stored in records

   record scan_env(subject, pos)

   procedure bscan(expr1)
      local outer_env

      outer_env := scan_env(&subject, &pos)
      &subject := expr1
      &pos := 1
      suspend outer_env   # pass saved environment to escan
      &subject = outer_env.subject
      &pos = outer_env.pos
      fail  # resume expression for expr1
   end

   procedure escan(outer_env, expr2)
      local inner_env

      inner_env := scan_env(&subject, &pos)
      &subject = outer_env.subject
      &pos = outer_env.pos
      suspend expr2
      outer_env.subject = &subject
      outer_env.pos = &pos
      &subject = inner_env.subject
      &pos = inner_env.pos
      fail   # resume expression for expr2
   end

(disclaimer: these routines are based on code that works, but I
typed them in without trying to translate or run them)

This bscan-escan expression works well if you are using a preprocess
(for example, one bassed on the variant translator system) to translate
from a notation that looks like a control structure, but looks
rather odd if you code it directly. However, it may give you some
ideas.

The only way I know of to solve the "break" problem is to use something
more primative than suspend (break just wipes out suspened generators
without considering that something more might need to be done), but you
can only do this in the run-time system routines written in C, not within
Icon itself.

  Ken Walker / Computer Science Dept / Univ of Arizona / Tucson, AZ 85721
  +1 602 621-4324  kwalker@cs.arizona.edu {uunet|allegra|noao}!arizona!kwalker