Re: [Chicken-users] s11n + rpc

[Sean Doran]

On 12 Nov, 2005, at 13:58, felix winkelmann wrote:

> I'm not sure how to proceed, though.

One approach would be to translate the binary output into a tagged/ 
marked-up format like *gasp* XML...

A sender would do something like write a tagged/marked-up (number- 
>string n).  What a receiver does with such a number too large or  
too small for its native fixnum is up to it.

#;1> (string->number "230984203849023894")
2.30984203849024e+17
#;2> (use numbers)
; loading /usr/local/lib/chicken/numbers.scm ...
; loading /usr/local/lib/chicken/numbers-base.so ...
#;3>  (string->number "230984203849023894")
230984203849023894

On the other hand, a markup language could tell the receiver that it  
must maintain a certain precision or fail.   Who knows what could  
evolve?

> Platform-independent serialization opens up a lot of possibilities  
> - on the other hand the serialization of procedures is highly  
> dependent of the compiler version and even optimization options...

You could keep the scheme source around and serialize that...   That  
presents lots of work for the receiver, but a memoization approach  
could avoid recompilation of already-received procedures.   This  
seems to call for a globally unique ID for the serialized procedure,  
and could be optimized with an ihave/sendme kind of dialogue.

s->r:  i have someprocguid
r->s:  i don't, so send it
s->r:  <serialized procedure>

receiver may save the serialized procedure and its own compiled  
version of it for future use, so that later:

s->r: i have someprocguid
r->s: i have it already

and thereby save the work of processing a serialized procedure.

This approach could be used for any serialized object, not just  
procedures.

Finally, using a globally unique id has advantages in terms of  
allowing for persistence and distribution.

        Sean.