Re: [Swig-user] Recursive functions in %extend: documentation should warn

[Andrey S. <dar...@ma...>]

Hendrik Maryns wrote:
> I was recently pointed to the %extend C_class mechanism, and loving it: 
> no more need for a self-constructed intermediate C file, I can directly 
> use the third-party C code now.
> 
> However, I encountered a problem there, which the documentation does not 
> warn for: invoking such a ‘virtual’ function recursively does not work, 
> one has to use the synthesized SWIG name in the recursive call.
> 
> E.g. I have a function to print a recursive struct:
> 
> struct mgTreeNode { /* tree node */
>     mA a; /* alphabet element */
>     struct mgTreeNode *left, *right; /* successors */
>     mgId id; /* state space id */
>     mgState state; /* automaton state */
> };
> 
> %extend mgTreeNode {
> 
>  char *printTreeNode(int labelLength) {
>   // reformat the label
>   char *label = calloc(labelLength+1, sizeof(char));
>   // …
>   // the daughters
>   char *leftDaughter = 0;
>   if ($self->left) {
>    leftDaughter = mgTreeNode_printTreeNode($self->left, labelLength);
>   }
>   char *rightDaughter = 0;
>   if ($self->right) {
>    rightDaughter = mgTreeNode_printTreeNode($self->right, labelLength);
>   }
>   // … print all stuff into buffer
>  return buffer;
>  }
> };
> 
> This works, but note the
>     mgTreeNode_printTreeNode($self->right,
> It would be much nicer if there were some kind of syntax to recursively 
> call the function.  I tried
>     $self->right.mgTreeNode_printTreeNode(
> but that doesn’t work.
> 
> So I guess this is an RFE, a suggestion to enhance the documentation, 
> and a call for help: is there a nicer way to do this?

May be nicer way to perform this task is to form recursion in separate 
function and call it from %extend method, like this:

%{
static char* printTreeNodeRecursively( mgTreeNode* self, int labelLength ) {
	// ...
}
%}

%extend mgTreeNode {
	char* printTreeNode( int labelLength ) {
		return printTreeNodeRecursively( $self, labelLength );
	}
}

In this case you will not use names 'mangled' by SWIG.

-- 
WBR, Andrey Starodubtsev        mailto:dar...@ma...