Hello,

There are a multi-thread example using SSockets in FPC?

I'm trying to implement it*, but it's very hard and this work it's consuming all my time. The difficulty is to read and write at the same time. I'm almost certain that using SSockets, I would have to create one socket for writing and one for reading, separately by threads.

Thank you!

* https://bitbucket.org/silvioprog/tcpipcomp (this repo is temporary, and this all code can undergo drastic changes at any time)

2013/4/24 silvioprog

... like ScktComp of Delphi. (no Synapse, LNet or Indy please, I need to use native code)

what is the problem with using Synapse, LNet or Indy? Seen the questions you are asking I would suggest you use an existing library and stay away from writing your own socket library or start with learning the basics of socket programming.

SOCK_STREAM sockets are full duplex and you can read and write simultaneously on them, even without threading if you use non blocking mode. SSocket has a Property NonBlocking but I don't see any code that handles the EAGAIN or EWOULDBLOCK errors. That is probably left to the user of TSocketStream to deal with. So SSocket is a really thin layer that leaves the user with all the problems you are trying to solve.

Ludo

Why is that ?

Please define "read and write at the same time" first.

Strange remark. Synapse, lnet or Indy are native pascal code.

Michael.

silvioprog wrote

There are a multi-thread example using SSockets in FPC?

It should be no problem at all to use sockets in Threads (native or via lnet or Synapse.

Of course if using threads, you will see the problem of inter-thread communication and/or tread-to-mainthread signaling.

For the later (more likely to be necessary) you can easily use TThread.Synchronize(), TThread.Queue() and QueueAsyncCall().

These will work out of the box with the WidgetTypes "Windows" and "KDE", and supposedly "GTK" and some Mac-flavors, but not with some of the others.

-Michael

silvioprog wrote

(no Synapse, LNet or Indy please, I need to use native code)

Indy does provide support for threading (as does AsyncPro, the Socket flavor of which could be of interest here). But I don't know if it's easy to use Indy with Lazarus, and I know that there is no Lazarus port for AsyncPro yet (even though it is open Source and especially the COM-Port flavor is requested rather often).

-Michael

2013/4/25 Ludo Brands

There no problem. Zeos has various features, auto-commit, script-executor, monitor etc.., But I prefer using SQLdb even when SQLdb was starting. :)

I'm not trying to write a library. I'm trying to solve a problem but using native units of Free Pascal (I miss a SktComp in Free Pascal, Delphi has it and works fine). After I solving my problem, I want to send the sources to be distributed in the samples of Free Pascal or Lazarus, so I'm avoiding third-party libraries.

SOCK_STREAM sockets are full duplex and you can read and write

Very interesting this information. I'll study more and try to solve this problem.

Thank you very much Ludo! :)

2013/4/25 Michael Van Canneyt

Why is that ? Please define "read and write at the same time" first.

Sorry, my English is bad. ^^

So, a small example (pseudo code):

procedure TMyThread1.Execute;
begin
  while not Terminated do
  begin
    GlobalSocket.Write ...
  end;
end;
procedure TMyThread2.Execute;
begin
  while not Terminated do
  begin
    GlobalSocket.Read ...
  end;
end;
MyThread1.Start;
MyThread2.Start;

Thank you!

Yes, I love Synapse, but it is not distributed with Free Pascal. I think the Free Pascal need to distribute something like SktComp( http://pastebin.com/QBZU89cy) of Delphi. SSockets is not multi-threaded.

2013/4/25 silvioprog

... and I'm trying to por SktComp to Free Pascal, but I do not know if the license of Delphi allow it. I ported only for studies.

2013/4/25 Michael Schnell

Indy has a feature (OnDisconnect property) that notifies you when a client is not online. I find this feature very interesting, and I intend to implement it with threads. But I seen that I have much work to do. :)

silvioprog wrote

Indy has a feature (OnDisconnect property) that notifies you when a client is not online.

Nope.

It notifies you when the other site actively and successfully does a disconnect and your site correctly understands the notification. Not correctly communicated aborts stay undetected.

-Michael

2013/4/25 Michael Schnell

I'll write an example for you to test, please wait ...

Maybe I can then update this old page? :)
http://www.brenemanlabs.com/iConnect.htm

2013/4/25 silvioprog

In attached.

Yes, OnDisconnect, it works exactly as you explained. But I also have the OnError event, which is useful when the client is terminated by an error (eg. killing the client via Ctrl+Alt+Del). :)

2013/4/25 Paul Breneman

hehe... I'll read it ...

Very nice article:

http://www.bastisoft.de/programmierung/pascal/pasinet.html

Outdated, but very very nice. :)

2013/4/25 silvioprog

in...:

repeat
   put all sockets currently connected in the read set;
   put the listening socket in the read set;
   Select(read set, no other sets, no timeout);
   respond to all sockets that are left in the read set;
until false;

:o So I need of "Select" function. If I am not mistaken it is not being used in ssockets. I'll to do it manually. I will implement it on my own.

Well, ssockets is very low level.

fpselect() is mostly useful when you combine multiple file it can be done, IMHO it makes little sense to create a select() method on the socket stream, since it would only have one socket handle.

You can perfectly do a fpselect() with the handles of the streams that ssockets uses.

Michael.

2013/4/25 Michael Van Canneyt

Nice.

The socket of Delphi and Synapse implements WaitForData:

function TWinSocketStream.WaitForData(Timeout: Longint): Boolean;
var
  FDSet: TFDSet;
  TimeVal: TTimeVal;
begin
  TimeVal.tv_sec := Timeout div 1000;
  TimeVal.tv_usec := (Timeout mod 1000) * 1000;
  FD_ZERO(FDSet);
  FD_SET(FSocket.SocketHandle, FDSet);
  Result := select(0, @FDSet, nil, nil, @TimeVal) > 0;
end;

I do not know where it would be used, it's just an observation hehe...

Yes, but the above is what I mean with 'it does not make much sense'.

I can easily add this - in fact I will do so, but with multiple sockets it is better to combine the waitfordata()

Michael.

In blocking mode the select is vital if you want to do simultaneous read and write on the same socket in the same thread. Otherwise you are simply blocking the thread in a recv when nothing comes in or in a send when the socket buffer is not empty. Note that select also returns when an exception occurs.

Ludo

I am aware of all this.
My only 'objection' is that waitfordata() will operate on a single socket only. Good for single client, bad for server.

Michael.

silvioprog wrote

Yes, OnDisconnect, it works exactly as you explained. But I also have the OnError event, which is useful when the client is terminated by an error (eg. killing the client via Ctrl+Alt+Del). :)

It might be able to detect a certain kind of errors.

Nonetheless, errors only arise when in fact communication is running. A socket stays open forever, if no communication is executed at all, thus a broken link in between stays undetected.

To detect a broken link you need to do a regular life check communication.

-Michael

In our previous episode, Michael Van Canneyt said:

How many open sockets must you have to qualify as a server?

My message is meant for the OP who wants to do simultaneous read/write on the socket in one thread and doesn't know what the use of select is. The 'it does not make much sense' is in that context misleading the OP and some more explanation was needed;)

Ludo

Well, you can't get more than ~ 65000 so lets go for half of that :-)

Michael.

In our previous episode, Michael Van Canneyt said:

I maybe have cut explanation short a bit, but I meant that not all servers are internet facing high load ones, where single thread per connection is a mortal sin.

I understood that. No problem.

Like I said, I'll add a WaitForData() or Select(read,write,exception) or something similar.

Just trying to find a 'nice' solution for multiple sockets. Probably a "class function select (array of socketstream)"

Michael.

2013/4/26 Ludo Brands

Exactly. In my small sample, when no buffer is received from server, the process stays blocked on:

https://bitbucket.org/silvioprog/tcpipcomp/src/f42be14db1e74d750fa88d6b4554b209145ff589/src/tcpipclient.pas?at=master#cl-189

2013/4/26 Michael Schnell

Pefectly. I'm trying to implement it. :)

2013/4/26 Ludo Brands

Ludo, a small question....

So, In loop of my thread, to the process doesn't stay blocked in recv, I need to use select in client too?

If your socket is in blocking mode and you don't want your thread to block, yes. Server or client, there is no difference. The same for send. When you have a small block to send that fits in the socket buffer and the socket buffer is empty, send will return immediately. But if you have more data to send, your second send will block until the first send is completed.
The same select can be used to return when the recv buffer is not empty or the send buffer empty or when an exception occurs. See http://linux.die.net/man/2/select

Ludo

silvioprog wrote

Select in socket of Delphi 2007 (a friend sent it to me: http://www.sendspace.com/file/06ev02):

Gaahhh!!! Don't simply send Delphi code to this list! Did you ever heard of stuff like clean room reverse engineeering?!

Regards,
Sven

2013/4/27 Sven Barth

Oooooooooopppss, sorry. I'll delete it. :X

you can't delete it from everywhere...

eg: systems that download this _mailing_list_ as separate emails into whatever mail reading tool is used...

2013/4/27 waldo kitty

From the SendSpace yes. Try to download it now. ;)

what "SendSpace"?? i've never heard of it... the code came in /inside/ your posting... [time passes] there was a link to the code in your post, yes... but you also posted the actual code, too...[/time passes]

my example explicitly points to local storage in a email client, as well... if you or anyone can delete posts from my local email storage from remote, i'd definitely like to know about it :P

Too late, gmane mail to news already stored it too.

L.

What? The link that shows 404 page not found? Or the actual content?

Your post to the mailing list, now is stored in NNTP system (at least in gmane)...

L.

I understand, but the problematic part was an upload to a third party site which has been deleted long since. So even browsing gmane/other archives, there's nothing left to see...

Reinier Olislagers wrote

I understand, but the problematic part was an upload to a third party site which has been deleted long since. So even browsing gmane/other archives, there's nothing left to see...

He also posted one of the Delphi functions to the list itself:

Jonas

Ah :(

the original message had a link and the actual code... deleting the link is one thing but the code is already distributed around the world and cannot be deleted by the OP...

2013/4/29 Jonas Maebe

Well, I messed up badly, sorry again.

I don't know if that would solve the problem, but if they prefer, we can delete me from the FPC lists, because I have no intention to piracy. Will be discretion of yours guys. x(

speaking only for myself: i think you have learned the lesson and there's no reason to kick you out of these lists...

what others do to clean up the mistake on their systems is up to them. of course... there may be a method to have the article cancelled from gmane and the other news servers that propagate the list in news format... the question is if the cancelmsg will propagate thru those news servers and be honored...

silvioprog wrote

I don't know if that would solve the problem, but if they prefer, we can delete me from the FPC lists, because I have no intention to piracy.

I know you don't, and there's definitely no way you will be removed from the list because you simply made a mistake.

Jonas

I guess the devs don't use gmane so they can just cat /dev/urandom > the mailbox file with the offending code ;) 3 times ;) interleaved with cat /dev/zero ;) same for the owner of the forum server.... if it got there.

as for the rest of us, we should be fine as long as Silvio's answer won't pop up or be referenced anywhere, so e.g. google won't index it somewhere visible... which probably is why we should end this thread immediately ;)

L.