NetNews Usenet Archive 1992 #31/NN_1992_31.iso/spool/comp/lang/scheme/2815

Internet Message Format | 1992-12-22 | 3.4 KB

Path: sparky!uunet!paladin.american.edu!gatech!usenet.ins.cwru.edu!agate!anarres.CS.Berkeley.EDU!bh
From: [e-mail] (Brian Harvey)
Newsgroups: comp.lang.scheme
Subject: Re: Spread of Scheme and SICP?
Date: 22 Dec 1992 20:16:46 GMT
Organization: University of California, Berkeley
Lines: 50
Message-ID: <[e-mail]>
References: <[e-mail]>
NNTP-Posting-Host: anarres.cs.berkeley.edu

[e-mail] (Terry Kaufman) writes:
>It would be a mistake to discourage high school teachers from using
>Scheme.  Besides as it continues to grow in popularity at the university
>level, Scheme has a much better chance than Logo to displace BASIC and
>Pascal in high schools.

I have mixed feelings about this.  As far as the core of the two languages is
concerned, I agree that Scheme has both easier-to-understand [although maybe
*initially* more intimidating] syntax and greater expressive power [i.e.,
lambda].  But two things work in Logo's favor, and I think at least one of
them should be addressed in R(N+1)RS:

1.  Debugging is easier in Logo, for a beginner, for two reasons:  (a) the
default behavior on an error is *not* to put you in some screwy debugging
sub-language that you don't understand; and (b) dynamic scope makes it very
easy to ask for all the relevant variables, without having to know about
frames, stacks, etc.

This is the one I think could be addressed in the standard.  I'd like it to
be standard behavior that you have to ask for a debugger before you get one.
Going along with that, the error message you see should be more complete,
telling you the expression that failed and the name (if any) of the procedure
in which it occurred.  [I know that that's hard to get right in a tail
recursive interpreter, but I can do it in Berkeley Logo, so it can be done!]
That would take care of part (a).

For part (b), I'd like to see a standard notation for asking for something
in a different frame.  I'm thinking along the lines of procedure:variable
to ask for some variable in the closest dynamically enclosing frame for an
invocation of some procedure.  Perhaps you might also be able to say
variable:3 for some variable three procedure calls ago, although that's
also hard with tail calling.  And I know I'm overloading the colon, but it's
not the notation I'm trying to design, just the meaning.  The point is that
the debugger should be a plain old Scheme REPL but allowing this extended
notation for variables in other frames without having to switch explicitly.

[Actually it wouldn't bother me in the least if the debugger just simulated
dynamic scope by letting you just say the variable name and, if there's no
such variable in the current environment, looking outward dynamically until
it finds one.  But I know that would be unpopular so I'm compromising.]

2.  Logo's word data type, which encompasses symbols, numbers, strings, and
characters, is incredibly convenient compared to all the x->y procedures you
have to hassle with in Scheme.  I understand that the Scheme way is more
efficient but Logo is great for a beginner who wants to find the plural of
a noun, or something like that.  [My colleague Matt Wright and I are about
to publish an introductory Scheme text, for people who aren't ready for
SICP yet, in which we use a version of Logo's data types implemented in
standard Scheme.  Since it *can* be done that way, I'm not so insistent on
making this be Scheme's standard behavior as for point #1.]