On Mon, 2007-11-26 at 17:45 +1100, skaller wrote:
> On Sun, 2007-11-25 at 18:24 -0800, Erick Tryzelaar wrote:
> > We could probably remove the ability to do 1i.
> >
> > What about just having a scoped syntax extension that allows you to
> > write "1.0 + 1.0i" in the complex module that isn't automatically
> > opened? For quaternions, you could add another that added support for
> > 1.0j and 1.0k.
> Just to make it clear: lexing is done first in a pass on a file,
> before any syntax analysis is done. That's why new tokens must
> be introduced with #preprocessor directives, and are scoped to
> the logical file (i.e. the physical file + #imports).
Just to add a note: new tokens can be created only in fixed
categories. Identifiers can be made 'user keywords', which
is a single token distinct from the usual NAME -- Dypgen
can detect the attribute dynamically. In addition, plain
identifiers can be checked without making them a user token.
This speeds up parsing by reducing the potential for ambiguity.
Sequences of special $%^&* characters and identifiers can also
be made into tokens. For the special characters the lexing really
IS interpreted by the lexer using a mini-lexer which is dynamic.
[It looks in a table for the longest match, removes that from
the input, then looks again for the next match, until the input
string is exhausted .. the table is managed by #infix etc
directives]
Dypgen can detect the attribute of the single user symbol here.
But still, none of this allows new NUMERIC or STRING kinds to
be introduced.
So the only way to handle 1.0i dependent on whether 'complex'
feature is enabled would be to pre-program it in one of two
ways and chose between them with some kind of preprocessor switch.
The point being the lexer would have to have builtin capability
to handle it anyhow.
Another method may be to simply consider i,j,k as constants in
a module of a special type. Then code like
1.0 i
might work, given an
const i : sqrt_minus_1 = "throw up";
fun apply (a:double, b:sqrt_minus_1) => fcomplex (0.0,a);
This is a neat trick. Note that 'i' is not a legal value in itself
in this formulation.
Note also this 'trick' is not so much a hack as you might think!
In fact there is a deep theory of field extensions developed by
the famous group theorist Galois, which basically corresponds
to this trick of taking a structure and 'adding' an abstract
symbol. Interestingly we have here a variable of the type,
yet it has no values! (Any attempt at evaluation would
lead to throwing up when compiling the C++ .. :)
--
John Skaller <skaller at users dot sf dot net>
Felix, successor to C++: http://felix.sf.net
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Felix-language mailing list
Felix-language@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/felix-language
