What should the development priorities for Felix be?
I have a huge list of things to do, including:
* library enhancements
* semantics upgrades
Semantics upgrades:
* fix addressing. At present, you can take the address of any expression
except a "val". The idea is that you can only take the address of:
(a) a C function marked "lvalue"
(b) a var
(c) a component of a var struct/tuple/record, recursively
In addition, addition to pointers should be fixed somehow: yet to
be designed how to do this!
* fix dot.operator
At present, structure/record/tuple component lookup is special, if that doesn't
work,
then we try reverse application. This also fails if the function is n-ary, i.e.
curried for
multiple arguments, and any ambiguity is not resolved on the first argument:
fun f(x:int) (y:int) => ...
fun f(x:in) (y:float)=>
println$ f 1 2.0; // ok
println$ 1.f 2.0; // fails
The idea is that
x.field
should be changed to
field x
by normal reverse application, and then the lookup work because x is a
structure.
Also, if x is a pointer to a structure we want
field (&x)
to return a pointer to the component. This is necessary so
(&x).field <- 42;
works. Note also this means that
x.2
will mean
2 x // third component of x
Since &(x.field) isn't allowed by the previous addressing fix,
you can't write:
x.field = 42;
unless we do something about it, so we can just define
&(x.field) --> (&x).field
as a parser or early desugaring hack (recursively). Then we get
field (&x)
as the interpretation. So in effect every struct defines two projections:
field: X -> t
field: &X -> &t
This means:
2: A -> t // third component of tuple or array
2: &A -> &t // pointer to third component
For an array:
a.expr --> (expr) a
(&a).expr -> (expr) a
In particular note the second formula means pointer arithmetic!
Comes automatically! I.e.
val n = 2;
n 2 == n.2 == n.[2]
*((&n) 2) == *((&n).2) == n.2
in particular:
val p = (&n).2;
val q = p + 1; // add 1 to pointer, same as C
This only works for array[T,N] in a variable.
It also isn't safe! Indexing varrays is safe.
We can make pointers into varrays safe too, with some
fierce hackery: the GC knows the head pointer and length.
* Make
inherit[U] X[U];
work. At present it goes into an infinite loop if I enable it
* Make class new work.
The code for this is done, it just isn't type safe and there's no
syntax for it. Class constructs a new object on the heap based
on a constructor, like C++, and, typically, delegated to C++.
The existing "new" operator does not do this, it *clones*
an object, i.e. uses the copy constructor.
The problem is primarily syntax:
new class X (1);
and also for the constructors:
constructor X: int = "X($1);
Note that "ctor" is NOT a type constructor, it's a value constructor,
that is, "ctor" are really factory functions or conversions, not
constructors ;( Maybe I should rename them all (gasp!) cvrt:
cvrt int: long = "(int)$1";
cast int: long = (int)$1"; // Argghhh
--
john skaller
skal...@users.sourceforge.net
http://felix-lang.org
------------------------------------------------------------------------------
Try before you buy = See our experts in action!
The most comprehensive online learning library for Microsoft developers
is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3,
Metro Style Apps, more. Free future releases when you subscribe now!
http://p.sf.net/sfu/learndevnow-dev2
_______________________________________________
Felix-language mailing list
Felix-language@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/felix-language
