gmane.lisp.cmucl.devel

Re: compare networking API on different CLs

Mike McDonald — 2008-07-24T16:30:07

I have no problems with bind-inet-socket and create-inet-socket. I still don't see the value in adding bind to connect-to-inet-socket. But the manner in which it has been added won't cause problems for those who don't care for it. So I vote for it, check the changes in. Mike McDonald mikemac< at >mikemac.com

Re: compare networking API on different CLs

Raymond Toy (RT/EUS — 2008-07-24T15:12:27

Chun> Right, you got it:) >> >> >> Although I don't do networking, I have heard usocket is a good socket >> library. Providing the necessary hooks so usocket can do what it >> needs to do would be good too. Chun> Thank you, very much! Well, I'd still like to hear other opinions. :-) Another option would be to port, say, sb-bsd-sockets. Or maybe provide a lower-level interface in addition to the high-level interface, with the high-level interface using the low-level. Ray

Re: compare networking API on different CLs

Chun Tian (binghe — 2008-07-24T14:37:06

Hi, Raymond 在 2008-7-24，下午9:58， Raymond Toy (RT/EUS) 写道： Right, you got it:) Thank you, very much! Regards, Chun Tian (binghe)

Re: compare networking API on different CLs

Raymond Toy (RT/EUS — 2008-07-24T13:58:09

Chun> 在 2008-7-24，下午7:01， Alex Goncharov 写道： Chun> Any "small extension" is to post-design CMUCL API, and why I do this Chun> is because CMUCL's networking API is incomplete. I don't know who Chun> design it, but if I can talk with this guy, I'll say: "not so good as Chun> other parts of CMUCL" The file says it was written by William Lott. It probably hasn't really changed much since it was written in 1992. (William Lott appears to have written a lot of the other parts of CMUCL too.) >> >> >> I don't use this kind of code and shouldn't be influenced by it. But >> I do care deeply about CMUCL stability, code organization and >> philosophy. Making CMUCL like others is not what I am longing for. Chun> I care about CMUCL stability too. If I'm talking about port the whole Chun> SB-BSD-SOCKETS package into CMUCL, I'm crasy, now I'm just trying to Chun> fix something what last CMUCL designer leave us to do. Chun> So, let's see other

Re: compare networking API on different CLs

Chun Tian (binghe — 2008-07-24T11:18:53

在 2008-7-24，下午7:01， Alex Goncharov 写道： Just add a encapsulate bind(2) cannot work with CMUCL's exist API. You must bind a socket *before* do other things to it. I think I have explained the reason in last post. I just follow CMUCL's exist API to make my changes. Any "small extension" is to post-design CMUCL API, and why I do this is because CMUCL's networking API is incomplete. I don't know who design it, but if I can talk with this guy, I'll say: "not so good as other parts of CMUCL" I care about CMUCL stability too. If I'm talking about port the whole SB-BSD-SOCKETS package into CMUCL, I'm crasy, now I'm just trying to fix something what last CMUCL designer leave us to do. So, let's see other post and idea, and let rtoy decide whether or not accept it. Regards, Chun Tian (binghe)

Re: compare networking API on different CLs

Alex Goncharov — 2008-07-24T11:01:22

,--- You/Chun (Thu, 24 Jul 2008 11:46:43 +0800) ----* | Yes, I do encapsulate *two* system calls, connect(2) and bind(2), in | CREATE-INET-SOCKET, one because there is *no* API in CMUCL which can | only do a bind(2)!!! If you do not give CREATE-INET-SOCKET local-host/ | local-port argument, the bind(2) will not be called. So, why not to suggest a simpler patch then -- encapsulating bind(2) only? | I have researched the networking API in five CLs (cmucl, sbcl, | openmcl, allegro, and lispworks) and I must say CMUCL has the most | poor API here. | CMUCL's networking API is quiet different from SBCL, but something | like LispWorks: It's probably not a bad comparison... | In SBCL, to do networking, (make-instance 'INET-SOCKET) first, it can | be TCP or UDP socket, then you may use SOCKET-BIND to bind it to local | port, or use SOCKET-CONNECT to call connect() on this socket. Now you | can see the difference: SBCL's SOCKET-CONNECT has argument as socket, | not host/port in CMUCL's CONNECT-TO-INE

Re: compare networking API on different CLs

Chun Tian (binghe — 2008-07-24T03:46:43

Hi, Alex Yes, I do encapsulate *two* system calls, connect(2) and bind(2), in CREATE-INET-SOCKET, one because there is *no* API in CMUCL which can only do a bind(2)!!! If you do not give CREATE-INET-SOCKET local-host/ local-port argument, the bind(2) will not be called. I have researched the networking API in five CLs (cmucl, sbcl, openmcl, allegro, and lispworks) and I must say CMUCL has the most poor API here. CMUCL's networking API is quiet different from SBCL, but something like LispWorks: In SBCL, to do networking, (make-instance 'INET-SOCKET) first, it can be TCP or UDP socket, then you may use SOCKET-BIND to bind it to local port, or use SOCKET-CONNECT to call connect() on this socket. Now you can see the difference: SBCL's SOCKET-CONNECT has argument as socket, not host/port in CMUCL's CONNECT-TO-INET-SOCKET. This means CMUCL has a high-level API than SBCL, but its high-level API lack of necessary arguments. LispWorks' networking API is quite like CMUCL (plus my patch), L

Re: [patch] Networking: bind to local-host/local-port when connect

Erik Huelsmann — 2008-07-23T18:34:25

Ah. Ok. Sorry for the misunderstanding. However, the only 2 implementations which prevent me from implementing this feature cross platform are CMU and CLISP. All others allow this action. This is one reason for 'wanting' it. Bye, Erik.

Re: [patch] Networking: bind to local-host/local-port when connect

Erik Huelsmann — 2008-07-23T18:20:44

It's not uncommon for servers to provide this functionality. That way the admin can limit the accepted input to the 'localhost' interface for local testing, for example. It's standard functionality in Apache, for example. I think it's standard in (x)inetd too. Bye, Erik.

Re: [patch] Networking: bind to local-host/local-port when connect

Chun Tian (binghe — 2008-07-23T18:34:46

在 2008-7-24，上午2:40， Mike McDonald 写道： For client to choose interface, I've said this in last post.

Re: [patch] Networking: bind to local-host/local-port when connect

Mike McDonald — 2008-07-23T18:40:03

I meant in CONNECT-TO-INET-SOCKET. Yes, servers sometimes want to bind() the socket to a specific interface. I can't figure out why a client would want to do that. Mike McDonald mikemac< at >mikemac.com

UDP server+client sample code

Chun Tian (binghe — 2008-07-23T15:34:32

Hi, CMUCL & rtoy I wrote a sample code for UDP networking test on CMUCL, based on my last patch. The source code in attach has four parts: Part 1 is my loadable patch to CMUCL's local-bind support, Part 2 is the WAIT-FOR-INPUT function from USOCKET project, Part 3 is my simple UDP server implementation, Part 4 is the UDP client test code. When running a UDP server, since I didn't use any multiprocessing, I must start two CMUCL (19e/darwin) to run test: On CMUCL A, I run a UDP echo server: * (compile-file "cmucl-udp-server" :load t) * (socket-server "localhost" 10000 #'identity) On CMUCL B, I run the client to test it: * (compile-file "cmucl-udp-server" :load t) * (socket-client-test) Then you should see text output on both side. CMUCL A: (server) * (socket-server "localhost" 10000 #'identity) I got something! I got 10 bytes from 2130706433:49621, it's #(1 2 3 4 5 6 7 8 9 10). I'll send this reply: #(1 2 3 4 5 6 7 8 9 10). 10 bytes sent. CMUCL B: (client) * (socket-client-test) I create

Re: [patch] Networking: bind to local-host/local-port when connect

Chun Tian (binghe — 2008-07-23T15:37:42

And another view: If you have more than one network interface in your machine, and you want to send your packets through specific interface, one way is to bind your socket to specific local address/port. Sometimes it's very important to have a way to choose the interface to use. --binghe 在 2008-7-23，下午11:01， Chun Tian (binghe) 写道：

Re: [patch] Networking: bind to local-host/local-port when connect

Mike McDonald — 2008-07-23T14:59:33

Out of curiousity, why would someone want to bind the local host and port to specific values? Mike McDonald mikemac< at >mikemac.com

Re: [patch] Networking: bind to local-host/local-port when connect

Chun Tian (binghe — 2008-07-23T15:01:14

Hi, Mike I think the biggest function of binding local host/port is to implement a UDP server. I'll show some code in half a hour. --binghe 在 2008-7-23，下午10:59， Mike McDonald 写道：

Re: [patch] Networking: bind to local-host/local-port when connect

Chun Tian (binghe — 2008-07-23T14:40:31

Update version of my patch in attach. Better and smaller. Current I do have a sample usage based on this patch, but it's usocket udp extension, and a portable UDP server. If you really want to see a pure-CMUCL example (a UDP server working with a UDP client, both in CMUCL), just give me some time to write one. --binghe 在 2008-7-23，下午9:26， Raymond Toy (RT/EUS) 写道：

Re: [patch] Networking: bind to local-host/local-port when connect

Raymond Toy (RT/EUS — 2008-07-23T13:26:06

Alex> [ Not sure cmucl-imp< at > will take a message from me -- a few were Alex> not posted in the past.] If that is still the case, please let me know. [snip] The patch is appended below. Ray (in-package :extensions) (ext:without-package-locks ;;; Added by Chun Tian (binghe) (defun bind-inet-socket (socket host port) "bind a socket to (local) host/port" (with-alien ((sockaddr inet-sockaddr)) (setf (slot sockaddr 'family) af-inet) (setf (slot sockaddr 'port) (htons port)) (setf (slot sockaddr 'addr) (htonl host)) (when (minusp (unix:unix-bind socket (alien-sap sockaddr) (alien-size inet-sockaddr :bytes))) (let ((errno (unix:unix-errno))) (unix:unix-close socket) (error 'socket-error :format-control "Error binding socket to port ~A: ~A" :format-arguments (list port (unix:get-unix-error-msg)) :errno errno))))) (defun connect-to-inet-socket (host port &optional (kind :stream) &key local-host local-port) "The ho

Re: [patch] Networking: bind to local-host/local-port when connect

Alex Goncharov — 2008-07-23T13:18:42

[ Not sure cmucl-imp< at > will take a message from me -- a few were not posted in the past.] ,--- You/Raymond (Wed, 23 Jul 2008 08:51:41 -0400) ----* | Chun Tian (binghe) wrote: | > CMUCL's networking API do not have standalone BIND function. And it's | > EXT:CONNECT-TO-INET-SOCKET and EXT:CREATE-INET-SOCKET function do not | > have any LOCAL-HOST/LOCAL-PORT argument. So a client socket is | > IMPOSSIBLE to bind to specific local host/port when CONNECT to remote | > address. This sounds terribly suspicious to me. I'd like to take a look at the patch but don't see any relevant message at http://news.gmane.org/gmane.lisp.cmucl.devel/ | > I wrote a loadable CMUCL patch, which create a new function | > BIND-INET-SOCKET, and add LOCAL-HOST, LOCAL-PORT keyword arguments | > to above two functions to support this feature. I think it's quite | > safe to merge it in next CMUCL version. I'd say, "Wait a second..." | The patch looks ok, but I'm not really competent to review the | changes since I never do

Re: [patch] Networking: bind to local-host/local-port when connect

Raymond Toy (RT/EUS — 2008-07-23T12:51:41

The patch looks ok, but I'm not really competent to review the changes since I never do network programming. Perhaps someone who is knows networking can say something. I do wish the functions didn't both have optional and keyword arguments, but the existing interface already does that. Too bad. Ray

Re: fixes for CMU User's Manual

Raymond Toy (RT/EUS — 2008-07-18T16:36:21

Thanks for the patch. It will be applied shortly. I don't know much about UDP networking and sendto, recvfrom, and shutdown, so I can't add any additional information to the documents. Ray

fixes for CMU User's Manual

Chun Tian (binghe — 2008-07-17T05:53:33

Hi, CMUCL / rtoy Recently I tried UDP networking in CMUCL then found some mistakes in CMU User's Manual: 1. Three functions INET-SENDTO, INET-RECVFROM, INET-SHUTDOWN is not external symbol of UNIX package but EXTENSIONS. 2. INET-SENDTO should have a ADDR argument. 3. INET-RECVFROM maybe need more document, or I don't know how to get receive data from a socket. Attach is my change to internet.tex, please review it. Regards, Chun Tian (binghe)