Hi,
>
> The thing that I wonder is whether it would be possible to embed Yacas
> at a lower level than via the parser. E.g. the Tcl embedding has
> Tcl_EvalObjEx() and friends, which operate on preparsed Tcl objects.
> Using this instead of Tcl_Eval() gives some efficiency gain, but more
> importantly, it avoids the "quoting hell" associated with the
> construction of the relevant command strings.
>
Interesting. You are not the first person asking this, actually.
I chose for the C API as it is nice and clean, no name mangling issues
while switching to a new gcc compiler, and it allows people to use it
from C also. Back then, I thought all people would want to do is send a
string and receive one back.
> Since both R and Yacas are "Lisp under the hood" languages (and Tcl
> too for that matter), it is a fairly good guess that their parse trees
> are similar, and that it would be a fairly simple task to translate
> back and forth between them. Well, between suitable subsets, at least.
> R's parse trees can in principle contain arbitrary R objects, but in
> this context, we might only need constants, names, operators, and
> function calls.
>
Very true. The code to chain up the linked list itself would not be very
little code, but we could iron out an initial API of course. It could
even
be through a C interface that returns void*.
Yacas can Already read a Lisp-style representation, to wit:
In> FromString("(foo(List a b))")LispRead()
Out> foo({a,b})
but that would not help you much as you mention you'd like to avoid
quoting hell.
From a C perspective, an API could look like this:
// Creating objects
void* yacas_create_atom(char* atom);
void* yacas_create_string(char* string);
void* yacas_create_number_from_string(char* string);
void* yacas_create_number_from_long(long number);
void* yacas_create_number_from_double(double number);
void* yacas_create_sublist(void* object);
// Linking linked lists
void yacas_link_objects(void* head, void* tail);
// executing an object
void* yacas_execute(void* object);
// pulling apart an object again
void* yacas_get_sublist(void* object);
char* yacas_get_atom(void* object);
char* yacas_get_string(void* object);
int yacas_object_is_string(void* object);
int yacas_object_is_atom(void* object);
int yacas_object_is_number(void* object);
int yacas_object_is_integer(void* object);
long yacas_get_long(void* object);
long yacas_get_double(void* object);
Note in the above list you would only really need get_sublist and
get_atom,
but using the others might be more efficient.
> It is probably not quite as easy as it sounds (is it ever?) but it
> could be worth it anyway. The main thing that it would require is an
> extended Yacas API which allows you to create Yacas objects from C
> code.
>
Apart from the above API, which I could create and would not take too
long,
there is of course the task of converting, in the sense that the
semantics
of R trees will be different from the semantics of the Yacas tree.
The other alternative is OpenMath, but that is just too verbose. I
don't like
xml-like schemes myself, as it is a very inefficient way of doing this.
Do you have cvs, and the possibility to compile under Windows?
Ayal
[[alternative HTML version deleted]]
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
