I've got a bunch of "legacy" Guile Scheme code that I want to get running in the Racket Scheme IDE. There appear to be enough differences to make this a non-trivial exercise. (My level of Scheme knowledge is the level to complete the The Little Schemer).
My question is:
What are the differences between Guile Scheme and Standard Scheme (in the Racket IDE)?
In light of these differences, in general, what are the steps I'll need to take to convert some Guile Scheme Code to standard Scheme?
Additional: (happy with divergence between Racket Scheme and R5RS/R6RS) - what I want is to get 'something' to run in the Racket IDE - rather than the Racket language.
If by "Standard Scheme (in the Racket IDE)," you mean the Racket language, i.e., what you get when you prefix your code with #lang racket, then the top four differences to look out for are:
a different module system
a different macro system (depending on how old your code is)
immutable cons-cells (unless you import mutable ones)
no one-armed ifs in Racket (use when)
To port code from Guile to Racket, find out which files are "at the bottom" of your dependencies, i.e., find the files that do not depend on other files.
Open such a file in Racket, add the line #lang racket at the top, and try to run it.
You will most likely encounter some "unbound identifier" errors.
If you are lucky, the function is present in Racket, but not included in the "racket" language. Search for the name in the Racket documentation, and if you find it, then use (require ...) to import the name into your program.
Then run the program again to find the next error.
Some function are named differently in Guile and Racket, so look up the name in the Guile documentation and see what it does. Then open the Racket documentation on the same subject, and see what it is called in Racket.
In some cases you may have to make bigger changes. If you can't find some piece of functionality in the Racket documentation, then try asking the mailing list. It could be that it simply has a different name, or that somebody implemented it and put it on PLaneT (thus it will no appear in the documentation until you have installed the package).
Example of importing srfi/1 into the R5RS language.
#lang r5rs
(#%require srfi/1)
(xcons 1 2)
Differences from R4RS code to modern Scheme?
One thing to look out for is that in R4RS the empty list '() counted as false, not it is interpreted as true.
See this question for more things to look out for:
Running SICP Pattern Matching Rule Based Substitution Code
See also this list of changes from the R5RS standard:
List of changes from R4RS to R5RS
Related
I wrote a program in Racket (the source code is in a .rkt file with #lang racket at the top). I also wrote a library in (mostly) portable R7RS Scheme. Can I use the library in the program in a clean way?
My goal is for the library to be widely portable between Scheme implementations (at least the R7RS-compliant ones, ideally others as well). There is a third-party R7RS shim for Racket but as far as I can tell it requires me to type #lang r7rs at the top of my source file. I presume this #lang directive would confuse Schemes other than Racket.
Can I put the core of my library in one or more portable .scm source files and then have one .rkt file with the #lang r7rs directive that tells Racket to include the portable files somehow? Does Racket understand some kind of library definition file such as the .sld used on snow-fort?
I tried to look all over the Racket documentation but I can't find this discussed anywhere. Nor did I find a general Scheme portability FAQ or best practices document.
I managed to mix Racket and R7RS code for real work and made an example of the technique on GitHub.
Here's a copy of the readme from that repo:
The Racket R7RS Shim
Racket doesn't ship with R7RS support. It's in the third-party package
r7rs by Alexis King: https://github.com/lexi-lambda/racket-r7rs
Despite not being an official part of Racket, it worked just fine for
me (I used a moderately complex library to do HTML parsing and wrote
some farily involved string processing and tree walking on top of it,
so this is definitely useful for real work).
You can install the shim via raco pkg install r7rs. Note also that
info.rkt lists r7rs in the dependencies, which you need for Heroku
and the like.
Modules
app -- a Racket application
lib -- an R7RS library used by app
sublib -- an R7RS library used by lib
What files the modules are made of
So app needs just one file, app.rkt, like any normal Racket
module.
But lib and sublib need 3 files each. lib.scm is the Scheme
code. lib.sld is the Scheme library definition. And lib.rkt is a
Racket wrapper for it. Technically you could combine lib.sld and
lib.scm into one file but it's cleaner to have them separate. You
could also copy all your Scheme code directly into lib.rkt but then
you can't import it into other Schemes.
Note that lib.scm doesn't have an (import ...) form at the top.
The imports are inside the define-library form in lib.sld. The
define-library form uses (include ...) to include the actual code
in lib.scm.
The job of lib.rkt is just to say #lang r7rs to Racket and then
include the Scheme stuff. It first needs to (import (scheme base))
so that we can use include and export. The included .sld files
import everything else from the Scheme standard that the library
needs.
Note that lib depends on sublib but sublib is not imported by
the define-library form in lib.sld. Instead, lib.rkt has to load
lib and all its dependencies: it contains (include "sublib.sld")
in addition to the obvious (include "lib.sld").
So lib.sld imports only stuff from the Scheme standard whereas
lib.rkt imports all our custom libraries. I had to resort to this
hack because I couldn't get the Racket module finder to find sublib
if I put it in the (define-library ...) imports. I didn't try hard
at all so there may well be a way to make it work.
Mutable vs immutable lists
Racket uses immutable cons cells (made by Racket's cons, satisfies
pair?) by default whereas R7RS uses mutable cons cells (made by
Racket's mcons, satisfies mpair?). That is, when you call cons
on the Scheme side, it actually makes something that looks to Racket
as if you had called mcons on the Racket side. A mutable cons means
you can use Scheme's set-car! and set-cdr! to alter it in place,
whereas the car and cdr of an immutable cons can't be changed after
the initial cons.
By default, Racket displays lists made out of mutable conses using
{curly braces} instead of (ordinary parentheses). This will bite
you when you pass lists over the R7RS--Racket boundary. You can print
using ordinary parentheses by changing the print-mpair-curly-braces
parameter but for many things it may be easier to convert your lists
(and trees) from mutable to immutable.
I don't know whether the Racket R7RS shim allows you to make immutable
conses on the Scheme side. It would be nice to have an option for
Scheme cons to make immutable conses (in that case set-car! and
set-cdr! would cause an error, which is fine for code using only
immutable data structures).
Where to find R7RS libraries
Lots of R7RS libraries are collected by Alex Shinn at
http://snow-fort.org/
Bottom line
The upshot of all this is that you can mix R7RS and Racket with a
little work and your codebase stays pretty clean (at least for simple
cases).
In Common Lisp, when I want to use different pieces of code depending on Common Lisp implementations, I can use *features* and the provided notation of #+ and #- to check the availability of a given feature and proceed accordingly. So for example (taken from Peter Seibel's PCL):
(defun foo ()
#+allegro (do-one-thing)
#+sbcl (do-another-thing)
#+clisp (something-else)
#+cmu (yet-another-version)
#-(or allegro sbcl clisp cmu) (error "Not implemented"))
Is anyone aware of a similar mechanism for Scheme? There are sometimes subtle differences between different implementations of Scheme, which, when you're trying to be portable, would be nice to abstract away. One such case that comes to my mind is Racket not providing mutable pairs by default. Instead of writing e.g. (set-cdr! lst '(1 2 3)) you would have to use set-mcdr! and only after you ran (require racket/mpair). Of course, such things could be abstracted by functions and/or macros, but I think the Common Lisp approach is neat in this aspect.
The closest thing there is, is cond-expand (aka SRFI 0), which is available on some Schemes but not others (Racket, for example, doesn't have it, and your code won't compile if you try to use it). For those Schemes that do have it, it looks like a cond form, except you test for booleans that tell you things about the compiler/interpreter. On some Schemes you can detect which Scheme you're running on, while on others you can only check for SRFIs:
(cond-expand (chicken
'bok-bok-bok!)
((and guile srfi-3432)
'this-guile-is-full-of-SRFI!)
(else
'(might be MIT Scheme, whose cond-expand only tests for SRFIs)))
A 4-year-old old post suggests that one might be able access the current-seconds and related functions in the r5rs language.
Here's why I ask: I'm a high school teacher new to Racket and we are using the r5rs language. I would like to introduce students to functions by starting with a function that needs no arguments to make sense. The example that occurs to me is minutes-past-the-hour. But I am ignorant of how to make those functions recognized in an r5rs program.
Thanks for any helpful advice.
First of all, why not use #lang racket instead of r5rs? Racket is very much built with education in mind. It even has various teaching languages for use with the How to Design Programs textbook (or its second edition, which is still being worked on).
Racket's implementation of R5RS is intentionally limited—it's not usually intended to be used for anything practical, since Racket itself has outgrown its Scheme roots. It can be useful as a teaching tool, but as you've seen, it doesn't include any special extensions (beyond a small set of internal forms).
If you're really interested in using R5RS Scheme, there exists an implementation of SRFI 19: Time Data Types and Procedures bundled with Racket. R5RS does not have a module system, so there is no formally-specified way of loading external libraries in pure Scheme. You'll need to use the Racket #%require extension to load the SRFI implementation:
(#%require srfi/19)
This will give you access to all the SRFI 19 functions and values.
You could also just include the functionality you want from Racket itself, since the languages are actually interoperable. To include current-seconds, you'd want to do something like this:
(#%require (only racket/base
current-seconds))
If you're going to do that, though, it seems almost pointless to use the r5rs language. Just use racket or racket/base instead.
Anyone know what is included in the racket/base language. I am looking to compare the racket/base namespace definitions with the R7RS draft to get a first hand idea of how divergent Racket is from Scheme.
The difference is going to be huge, just like any other practical Scheme implementation. To give you a rough idea, I see more than 1300 identifiers from racket/base, and racket goes up to over 2000. Also, note that comparing the names is not going to be the whole story -- there are additional differences, like the fact that Racket uses immutable pairs by default, and like the fact that you get guaranteed left-to-right evaluation order.
A quick way to get the lists yourself for the former would be to make sure that XREPL is installed (have your ~/.racketrc file contain (require xrepl)), then start Racket with racket/base as the REPL language, and use the ,ap (apropos) command with no arguments so it shows you all of the bound names:
$ racket -I racket/base
Welcome to Racket v5.2.1.
-> ,ap
To read about the names, you can use the ,doc command, or just use the search box in the docs.
According to The Racket Reference:
Unless otherwise noted, the bindings defined in this manual are exported by the racket/base and racket languages.
Unfortunately that is not terribly helpful since it is a large reference manual. But, since you are comparing R7RS to Racket, it may be useful to just browse through the whole reference to get an idea of what is available.
Background: I am new to scheme, and am using DrScheme to write my programs.
The following program outputs 12345 when I run the program as r5rs:
12345
However the following program outputs nothing (it's an r6rs program):
#!r6rs
(import (rnrs))
12345
That being said, I can get it to output 12345 by doing this:
#!r6rs
(import (rnrs))
(display 1235)
Is that something new with r6rs, where output only occurs when specifically specified using display? Or am I just doing something else wrong
This is a subtle issue that you're seeing here. In PLT, the preferred mode of operation is to write code in a module, where each module has a specification of the language it is written it. Usually, the default language is #lang scheme (and #! is short for #lang). In this language, the behavior is for all toplevel non-definition expressions to display their values (unless they're void -- as in the result of most side-effects). But the #lang r5rs and #lang r6rs don't do the same -- so these toplevel expressions are evaluated but never displayed.
The reason you did see some output with the R5RS language is that you didn't use it as a "module" (as in #lang r5rs), but instead used the specific R5RS "language level". This language level is more compatible to the R5RS, but for various subtle reasons this is not a good idea in general. Using #lang is therefore generally better, and if you want to save yourself some additional redundant headaches, it'll be easier if you stick with #lang scheme for now, and worry about standards later. (But YMMV, of course.)