Re: Adding dot-notation in an ISO conforming manner

On 12/10/2013 05:13 AM, Richard A. O'Keefe wrote:
> I seem to have missed something.
>
> Does this mean that if I do
>	write_canonical([a,b,c])
> in some other Prolog, getting
>	'.'(a, '.'(b, '.'(c, [])))
> and then try to read this in SWI Prolog,
> it will read as something *other* than [a,b,c]?
>
> Why is this thought to be a good idea?

It is not. It is a consequence though and not too bad. First of all,
SWI-Prolog can talk to itself and to other Prolog systems. It is just
that SWI-Prolog cannot read the write_canonical/1 output of other
systems. That is a pity of course. As a work around, there is an option
dotlists(true) to read_term/3 and write_term/3 which causes these to
pretend that lists are ./2 terms.

For canonical exchange between Prolog systems, using the dotlists(true)
option for reading should suffice. If you have non-Prolog readers you
might have to abandon write_canonical/1 and use write_term/2 with
dotlist(true), quoted(true) and ignore_ops(true). Or better, change the
reader to understand the list syntax. That will probably fix handling
long lists as well ...

> Oh well, since [] reads as the wrong kind of data
> structure anyway, it was already broken.
>
> But the *point* of write_canonical was to facilitate
> data interchange between Prologs.
>
> Changing the list constructors from '[]' and '.' is
> looking like a worse and worse idea.
>
> There isn't any problem with write_canonical reading
> and writing strings, but getting list reading wrong
> is, well, ...

There is a much bigger problem if the data representation relies on
strings not being lists. That problem would really upset exchange, while
the above problem requires merely changing a flag.

	Cheers --- Jan