Re: [Chicken-users] call Chicken Scheme from C and pass a bytevector

[Claude Marinier]

On Thu, 27 Jun 2013, Dan Leslie wrote:
> There's a section on accessing external objects that covers this sort of 
> thing:
> http://wiki.call-cc.org/man/4/Accessing%20external%20objects#returning-large-objects-or-chunks-of-memory-to-scheme

This addresses a different issue. The PCAP event handler in C calls the 
packet handler in Scheme. I am looking for a way to create saddr_o and 
daddr_o and give them to scheme like this.

  vector = C_alloc( C_SIZEOF_VECTOR(7) );
  vector_o = C_vector( &vector, 7,
    C_fix(ethtype),
    saddr_o, daddr_o,
    C_fix(ip_proto),
    C_fix(type), C_fix(code),
    C_fix(tot_len) );

  ret = process_packet( vector_o );  // call Scheme code

> It's possible to allocate C structures under the control of the Chicken GC:
> http://wiki.call-cc.org/allocating-c-structures-under-control-of-the-chicken-gc

Hum ... this has potential.

  struct ipv4_addr_struct {
    C_header tag;
    uint8_t *octets;
  };
  typedef struct ipv4_addr_struct ipv4_addr;

  static const C_header BTREE_TAG =
    ((sizeof(ipv4_addr) - sizeof(C_header)) /
    sizeof(C_word)) | C_BYTEVECTOR_TYPE;

This would be handled by the GC, right?

Would Scheme be able to convert or use this as a u8vector or would it be 
a blob which I could convert with blob->u8vector/shared ?

> However, you can also leave control up to the user rather than the GC and 
> they can use free:
> http://api.call-cc.org/doc/lolevel/free

I prefer to let the GC handle it.

> Personally, I like to allocate blobs and srfi-4 vectors in scheme and pass 
> them as parameters to C functions to be mutated.

Wrong direction: I call scheme from C.

> On 6/27/2013 10:07 AM, Claude Marinier wrote:
> > Hi,
> >
> > A function in pcap-interface.c calls Chicken Scheme. It builds a vector 
> > containing a bunch of things, e.g. C_fix(ethtype), and the source and 
> > destination addresses as vectors. The scheme code converts the address 
> > vectors to u8vector (u16vector for IPv6). This is a lot of work: just under 
> > 20% of CPU time is spent in vector->u8vector.
> >
> > If I could create a blob in C and pass it to scheme, the scheme code could 
> > use blob->u8vector/shared and blob->u16vector/shared which would achieve 
> > the same result with much less work.
> >
> > What is the recommended way to pass a blob to scheme from C?
> >
> > Is there a way to create a u8vector or u16vector directly in C?

Thank you.

--
Claude Marinier