Keyword syntax

Keyword syntax

[Marius Vollmer]

Hi,

I'm considering to change keywords are read or written by Guile, to
make the two operations more consistent with each other.

Right now, keywords are read as a single token and a part of that
token is made the keyword name, which is a symbol.

For example, a token can be "#:foo" (without the double quotes) and
the name of the keyword is the symbol named "foo".  Also, "#:12" gives
a keyword with a symbol as the name that itself has the name "12".
Now, "12" is not a valid way to write this symbol in Scheme, and there
for Guile writes the keyword #:12 in the following funny way:

    guile> #:12
    #:#{12}#

The "#{12}#" is the syntax to write a symbol with an arbitrary name.

However, reading this printed form gives a different keyword.

    guile> #:#{12}#
    #:#{\#{12}\#}#


I can see two ways to fix this: 1) changing the way keywords are read and
2) changing the way they are printed.

As to 1), the simplest change would be to just do the equivalent of
(symbol->keyword (read port)).  Thus, the name of a keyword is read as
a general Scheme datum, and is then validated to be a symbol and
converted to a keyword.

For 2), we would have to print keyword without using the symbol
printer and would have to explicitely deal with 'weird' keyword names
ourselves.

I have implemented 1) in CVS head as an experiment.  Below is a sample
session, showing the differences.

I like 1) more than 2) (by a small margin) although it is a more
radical change since keywords right now are defined to have symbols as
names anyway and because it is equivalent to doing

    (read-hash-extend #\: 
      (lambda (chr port)
        (symbol->keyword (read port))))

which I like because it is very straightforward.


    Previously

        guile> #:12
        #:#{12}#
        guile> #:#{12}#
        #:#{\#{12}\#}#
        guile> #:(a b c)
        #:#{}#
        ERROR: In expression (a b c):
               Unbound variable: a
        guile> #: foo
        #:#{}#
        ERROR: Unbound variable: foo

    With 1)

        guile> #:12
        ERROR: Wrong type (expecting symbol): 12
        guile> #:#{12}#
        #:#{12}#
        guile> #:(a b c)
        ERROR: Wrong type (expecting symbol): (a b c)
        guile> #: foo
        #:foo

Opinions?


_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Keyword syntax

[Roland Orre]

I like 1 more also. (I often use the non standard scheme way of
reading values with #something but I don't think that will
be a problem)

	Roland

On Mon, 2004-10-04 at 20:27, Marius Vollmer wrote:
> Hi,
> 
> I'm considering to change keywords are read or written by Guile, to
> make the two operations more consistent with each other.
> 
> Right now, keywords are read as a single token and a part of that
> token is made the keyword name, which is a symbol.
> 
> For example, a token can be "#:foo" (without the double quotes) and
> the name of the keyword is the symbol named "foo".  Also, "#:12" gives
> a keyword with a symbol as the name that itself has the name "12".
> Now, "12" is not a valid way to write this symbol in Scheme, and there
> for Guile writes the keyword #:12 in the following funny way:
> 
>     guile> #:12
>     #:#{12}#
> 
> The "#{12}#" is the syntax to write a symbol with an arbitrary name.
> 
> However, reading this printed form gives a different keyword.
> 
>     guile> #:#{12}#
>     #:#{\#{12}\#}#
> 
> 
> I can see two ways to fix this: 1) changing the way keywords are read and
> 2) changing the way they are printed.
> 
> As to 1), the simplest change would be to just do the equivalent of
> (symbol->keyword (read port)).  Thus, the name of a keyword is read as
> a general Scheme datum, and is then validated to be a symbol and
> converted to a keyword.
> 
> For 2), we would have to print keyword without using the symbol
> printer and would have to explicitely deal with 'weird' keyword names
> ourselves.
> 
> I have implemented 1) in CVS head as an experiment.  Below is a sample
> session, showing the differences.
> 
> I like 1) more than 2) (by a small margin) although it is a more
> radical change since keywords right now are defined to have symbols as
> names anyway and because it is equivalent to doing
> 
>     (read-hash-extend #\: 
>       (lambda (chr port)
>         (symbol->keyword (read port))))
> 
> which I like because it is very straightforward.
> 
> 
>     Previously
> 
>         guile> #:12
>         #:#{12}#
>         guile> #:#{12}#
>         #:#{\#{12}\#}#
>         guile> #:(a b c)
>         #:#{}#
>         ERROR: In expression (a b c):
>                Unbound variable: a
>         guile> #: foo
>         #:#{}#
>         ERROR: Unbound variable: foo
> 
>     With 1)
> 
>         guile> #:12
>         ERROR: Wrong type (expecting symbol): 12
>         guile> #:#{12}#
>         #:#{12}#
>         guile> #:(a b c)
>         ERROR: Wrong type (expecting symbol): (a b c)
>         guile> #: foo
>         #:foo
> 
> Opinions?
> 
> 
> _______________________________________________
> Guile-user mailing list
> Guile-user@gnu.org
> http://lists.gnu.org/mailman/listinfo/guile-user



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Keyword syntax

[Paul Jarc]

Marius Vollmer <marius.vollmer@uni-dortmund.de> wrote:
> As to 1), the simplest change would be to just do the equivalent of
> (symbol->keyword (read port)).

I like this way too.

>         guile> #:12
>         ERROR: Wrong type (expecting symbol): 12

You could add a bit of backward compatibility:
(let ((obj (read port)))
  (symbol->keyword
   (if (symbol? obj)
     obj
     (string->symbol (object->string obj)))))

But I'm not sure this would be worth it, or if there may be corner
cases where it produces a different keyword.


paul


_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Keyword syntax

[Marius Vollmer]

prj@po.cwru.edu (Paul Jarc) writes:

> Marius Vollmer <marius.vollmer@uni-dortmund.de> wrote:
>> As to 1), the simplest change would be to just do the equivalent of
>> (symbol->keyword (read port)).
>
> I like this way too.

Ok, I'll keep it that way then.

>>         guile> #:12
>>         ERROR: Wrong type (expecting symbol): 12
>
> You could add a bit of backward compatibility:
>
> (let ((obj (read port)))
>   (symbol->keyword
>    (if (symbol? obj)
>      obj
>      (string->symbol (object->string obj)))))
>
> But I'm not sure this would be worth it, or if there may be corner
> cases where it produces a different keyword.

This wouldn't be very much backwards compatible, I'm afraid, since
'read' is very much different from reading a token (with an internal
reader function).

We could keep more backwards compatability by continuing to read the
keyword name as a token, but only acception names that are proper
symbol names.  Should we?


_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user