Services & ResourcesWolfram ForumsMathGroup Archive

[Date Index] [Thread Index] [Author Index]

Re: raw TCP/IP socket communication in mathematica

• To: mathgroup at smc.vnet.net
• Subject: [mg17619] Re: raw TCP/IP socket communication in mathematica
• From: dreeves at flip.eecs.umich.edu (Daniel Reeves)
• Date: Tue, 18 May 1999 02:45:20 -0400
• References: <Pine.GSU.4.05.9905041712540.21049-100000@flip.eecs.umich.edu> <7grica$e6h@smc.vnet.net>
• Sender: owner-wri-mathgroup at wolfram.com

I took paulh at wolfram's advice, and implented an interface to unix sockets
using mathlink.

http://ai.eecs.umich.edu/people/dreeves/misc/math-sockets/

It includes a sample notebook that grabs an arbitrary webpage from the
internet and spits out the contents.  About 4 lines ("link to socket
library", "connect to web site", "ask for web page", "receive response").

I still think raw TCP/IP communication should be part of the kernel.  To
Jens's comment that it can't be because his windows box can't do it, I
have 2 responses:  1, his machine would fail to create any link to a
remote machine, "Raw" or otherwise.  2, Mathematica does not take a lowest
common denominator approach to operating system functionality, eg, piping
to external programs is supported in Mathematica even though it won't work
on some operating systems.

So again I'd like to urge Wolfram developers to consider including this
functionality in the kernel.  I'd like to see Mathematica become the
language of choice for things like "intelligent internet agents" (like
bidding agents for online auctions) and not having built-in ability to
send and receive data over the internet is a big enough hurdle to prevent
that from happening.  Compiling/porting/etc something like what I've done
is a huge headache.  But inside the kernel, this problem is already solved
-- there just needs to be an option to turn off enforcement of the
structured, mathlink-packet-based communication and use raw strings
instead.

PS: When I first posted my message about my MASH thing, I accidentally
didn't have the source code (mash.c) visible.  It's now there at
http://ai.eecs.umich.edu/people/dreeves/misc/mash/

Thanks,
Daniel

--    --    --    --    --    --    --    --    --    --    --    -- 
Daniel Reeves               http://ai.eecs.umich.edu/people/dreeves/

"'Artificial Intelligence' ceases to be 'intelligent' as soon as it's
actually implemented."  -- Uluc Saranli

On 6 May 1999, P.J. Hinton wrote:

> On Tue, 4 May 1999, Daniel Reeves wrote:
> 
> > I think the ability to send and retrieve arbitrary strings over TCP
> > sockets is very important.  The specific application I have in mind is
> > creating bidding agents that participate in an online auction (part of my 
> > research on artificial intelligence for ecommerce).  This type
> > of application is becoming more and more common and I think it's important
> > that Mathematica support communication with programs other than mathlink
> > compatible ones.  Bots that gather data on the web is another example of
> > why this would be necessary.
> > 
> > It should be straightforward to implement this in the kernel by having an
> > option for LinkConnect that says "Raw".  Then all LinkWrite's and
> > LinkRead's would send and receive plain strings.
> 
> Alternatively, you could build an installable MathLink binary that defines
> a top-level interface to your operating system's native socket API.  That
> would be a highly reusable component that could be launched whenver
> needed.
> 
> Below is an example of a top-level interface to the Unix system call
> uname(). You could create a MathLink template like this:
> 
> :Begin:
> :Function: myuname
> :Pattern: SystemInformation[]
> :Arguments: {Null}
> :ArgumentTypes: Manual
> :ReturnType: Manual
> :End:
> 
> and then the C code would look something like this:
> 
> #include <sys/utsname.h>
> #include "mathlink.h"
> 
> void myuname(void);
> 
> void myuname(){
>   struct utsname unamedata;
>   int retval;
> 
>   retval = uname(&unamedata);
>  
>   if(retval == 0){
>     MLPutFunction(stdlink, "List", 6);
>     MLPutString(stdlink, unamedata.sysname);
>     MLPutString(stdlink, unamedata.nodename);
>     MLPutString(stdlink, unamedata.release);
>     MLPutString(stdlink, unamedata.version);
>     MLPutString(stdlink, unamedata.machine);
>     MLPutString(stdlink, unamedata.domainname);
>    }
>   else{
>     MLPutSymbol(stdlink, "$Failed");
>    }
>  
>    return;
> }
> 
> int main(int argc, char *argv[]){
>   return MLMain(argc, argv);
> }
> 
> I can now get the result of uname from my installed function.
> 
> In[1]:= Install["myuname`"]
> 
> Out[1]= LinkObject['./myuname.exe', 1, 1]
> 
> In[2]:= LinkPatterns[%1]
> 
> Out[2]= {SystemInformation[]}
> 
> In[3]:= SystemInformation[]
> 
> Out[3]= {Linux, monon, 2.0.0, #1 Mon Jun 10 21:11:56 CDT 1996, i586,
> (none)}
> 
> --
> P.J. Hinton	
> Mathematica Programming Group		paulh at wolfram.com
> Wolfram Research, Inc.			http://www.wolfram.com/~paulh/
> 
> 

• Follow-Ups:
  • Re: Re: raw TCP/IP socket communication in mathematica
    • From: "Mitja Lakner" <mlakner@fagg.uni-lj.si>