🚀 go-pugleaf

RetroBBS NetNews Server

Inspired by RockSolid Light RIP Retro Guy

Thread View: borland.public.cppbuilder.internet.socket
3 messages
3 total messages Started by Carlos Moreno Mon, 21 Apr 2008 15:54
Trouble with TcpClient component
#73
Author: Carlos Moreno
Date: Mon, 21 Apr 2008 15:54
81 lines
4158 bytes
Hi,

I'm having major trouble and major frustration when trying to 
migrate an application from C++ Builder 6 to Builder 2007  (I 
have the trial version, so it may not be the latest release 2).

The short story / bottom line:  how can I emulate the non-
blocking functionality of TClientSocket, where I transmit 
whenever I need to transmit, and get an event whenever the 
socket received data from the server?

The long / detailed story:

My existing application uses sockets (TClientSocket) in 
non-blocking mode for *fully asynchronous* communications.

TClientSocket is not supported in C++B2007, and the closest 
thing seems to be TcpClient (which also exists in C++B6, so 
this could help with the transition).

With TClientSocket, I set it in non-blocking mode, and I do 
socket->Open() when the user activates the connection.  Works 
like a charm:  whenever the user transmits something, I receive 
an OnReceive (or was it OnRead?) event, to which I respond by 
calling socket->ReceiveBuf and read the received data.

I pretty much ignore the OnSend event, as that one simply tells 
me that I *can* transmit --- I simply transmit when the user 
triggers an action that requires transmitting.

Ok, I'm trying to emulate exactly the above --- the docs in 
C++B6 for this class/component are utterly useless;  and so are 
C++B2007's --- The entry TcpClient takes me to the documentation 
for TIdTcpClient --- Indy's tcp client socket, which does not 
look even remotely similar to TcpClient.

The little it says suggests that it works exactly the same; 
however, they don't.

When I call socket->Open(), I immediately receive an error 
(an OnError event, with error code WOULDBLOCK --- meaning 
that the call to connect returned immediately but the operation 
has not been completed....  Well DUH!   It's in non-blocking 
mode !!!  How could it be completed?  And how why would it tell 
the application that there was an error?

(needless to say that with TClientSocket, I get no error, and 
instead, I get an OnConnect event a fraction of a second after 
calling socket->Open() ....  Seems like the obviously correct 
thing to do)

The connection *does* succeed, but I never get an OnConnected 
event  (I know it succeeds for two reasons:  1) if I call 
socket->Open() again, I get an error ALREADYCONNECTED;  and 
2) if I now call socket->ReceiveBuf, I do read the server's 
greeting message --- the server transmits immediately upon the 
connection being accepted)

The thing is, I never get an OnReceive event (well, I do, but 
I do get it in response to my call to ReceiveBuf --- clearly, 
I need it to be the other way around;  I would only call 
ReceveiBuf when I'm notified, via an event, that the socket 
just received data);  I really really need this feature, since 
my application has other things to do and can not afford to do 
polling on the socket to see if there is data, nor set it in 
blocking mode.

Sure, I could make a separate thread and do polling or use the 
socket in blocking mode, but that would be a major restructuring 
of the application, so I'd *really* like to stay away from that 
option.

Any tips on what's going on or how to make this work?  Am I 
doing something wrong?  Is the functionality of these sockets 
hopelessly broken?  (in which case, suggestions?)

Thanks!

Carlos
--
Re: Trouble with TcpClient component
#74
Author: Remy Lebeau (TeamB)
Date: Mon, 21 Apr 2008 16:36
67 lines
3866 bytes
"Carlos Moreno" <moreno_h3gd@mochima.com> wrote in message 
news:480cb8b5$1@newsgroups.borland.com...

> The short story / bottom line:  how can I emulate the non-
> blocking functionality of TClientSocket, where I transmit
> whenever I need to transmit, and get an event whenever the
> socket received data from the server?

TClientSocket is deprecated, but it is still available in BCB 2007 (install 
the dclsockets100.bpl packet), and works the same as it did before.

> TClientSocket is not supported in C++B2007

Yes, it is.  IT is simply not installed by default anymore.

> and the closest thing seems to be TcpClient

TTcpClient is nowhere even close to TClientSocket.  They are very different. 
TTcpClient was created at a time when Borland was trying to write its own 
cross-platform socket classes.  Those classes were not implemented very 
well, though, and lack a LOT of functionality.  I strongly recommend not 
using them at all.  Stick with TClientSocket instead, or else use a 
third-party sockt library instead, such as Indy, ICS, or Synapse.

> I pretty much ignore the OnSend event

That is a mistake, especially since you were using a non-blocking socket. 
If you trying to pass too much data to TClientSocket at one time, 
SendBuf/Text() fails with a return code of -1 and WSAGetLastError() returns 
WSAEWOULDBLOCK.  In that situation, you need to wait for the OnSend event to 
fire before you can pass any further data to the socket - including 
re-sending the data that triggered the blockage in the first place!  A lot 
of TClientSocket (and TServerSocket) users ignore that fact when they 
shouldn't be.

> as that one simply tells me that I *can* transmit --- I simply transmit
> when the user triggers an action that requires transmitting.

But that is still not a guarantee that the socket will be able to accept 
everything you tell it to transmit.  If it doesn't, and if you do not handle 
it properly, then you will corrupt your communications with the other party.

> The little it says suggests that it works exactly the same;

TTcpClient does not work like TClientSocket, not even close.

> The thing is, I never get an OnReceive event

TTcpClient DOES NOT have asynchronous events at all!!!

> well, I do, but I do get it in response to my call to ReceiveBuf

Which is exactly how TTcpClient is coded to work.  You will NOT get the 
OnReceive event asynchronously!  You have to continuously invoke your own 
reading operations, and then TTcpClient will read whatever is available at 
the time.

> clearly, I need it to be the other way around;  I would only
> call ReceveiBuf when I'm notified, via an event, that the
> socket just received data)

There are no asynchronous notifications in TTcpClient.  Which is one of the 
reasons why I hate that component and never advise anyone to use it.


Gambit 
Re: Trouble with TcpClient component
#75
Author: Carlos Moreno
Date: Mon, 21 Apr 2008 16:55
55 lines
2689 bytes

Hi Remy!

>TClientSocket is deprecated, but it is still available in BCB 2007 (install 
>the dclsockets100.bpl packet), and works the same as it did before.

OH!!   That's actually good news, I guess...  (I'm somewhat 
having a flashback of having been told this long ago --- C++ 
Builder 2006, perhaps? ...  But I think at the time I did not 
have much reasons to want to migrate, so a default that did 
not directly support my existing application felt enough to 
decide not to migrate...  It is also conveivable that I'm 
just having a random deja-vu when reading your comment above)

>> and the closest thing seems to be TcpClient
>
>TTcpClient is nowhere even close to TClientSocket. 

From your description, I now completely agree with you... 
The thing is, it does have about the same properties, the 
same events, and the little docs available give descriptions 
that suggest that the functionality is indeed the same.

For instance, this is BC++B6's description of OnReceive for 
TTcpClient:

"Occurs after a socket receives data, but before it passes the 
data to the application 
    [ ... ]

Description
Write an OnReceive event handler to read from the socket 
connection ..."

I don't know, maybe I'm misinterpreting the above, but it did 
sound a lot like a description of how TClientSocket works! :-(

> I strongly recommend not using them at all. 

You definitely sold your case, no argument there!

>> I pretty much ignore the OnSend event
>
>That is a mistake [ ... ]

Well, I was simplifying a bit my post --- I do handle the 
possibility of overflowing the local buffer when transmitting. 
But good to hear the additional comments about the issue.


Thanks again!  Your message was extremely helpful!! 

Carlos
--
Thread Navigation

This is a paginated view of messages in the thread with full content displayed inline.

Messages are displayed in chronological order, with the original post highlighted in green.

Use pagination controls to navigate through all messages in large threads.

Back to All Threads