Problems with Socket>>isPeerAlive

Hi,

in the attachment there is a piece of code, which establishes a tcp
connection to a listening socket and reads all arriving data from there.

I thought i could use Socket>>isPeerAlive to test whether i can expect
further data from the prev. listening socket.

But then i shutdown the end point of the connection (the prev. listening
socket) isPeerAlive still signals true.

Is this wrong usage of isPeerAlive? Should i rather use Socket>>atEnd to
test whether the connection is still alive? Or something else?

Steps to reproduce:

  - open a listening socket on localhost port 1234
    (i use netcat: 'nc -l -p 1234')
  - start the Communicator.st like this:
    gst Communicator.st
  - ctrl-c/abort the netcat/listening process after Communicator
    connected
  - watch 'read hunk []' scrolling on the screen where you started gst.

Robin

--
and why is my signature missing?


_______________________________________________
help-smalltalk mailing list
[hidden email]
http://lists.gnu.org/mailman/listinfo/help-smalltalk

On Tue, Oct 31, 2006 at 12:43:47PM +0100, Robin Redeker wrote:
> Hi,
>
> in the attachment there is a piece of code, which establishes a tcp
> connection to a listening socket and reads all arriving data from there.

And as usual i forgot the attachment...

_______________________________________________
help-smalltalk mailing list
[hidden email]
http://lists.gnu.org/mailman/listinfo/help-smalltalk

Robin Redeker wrote:
> On Tue, Oct 31, 2006 at 12:43:47PM +0100, Robin Redeker wrote:
>  
>> Hi,
>>
>> in the attachment there is a piece of code, which establishes a tcp
>> connection to a listening socket and reads all arriving data from there.
>>    
Yes, the SIGIO arrives but events.c misses it because a POLLHUP
condition (peer dead) does not increment the return value of poll.

Fixing this in turn triggers a latent bug in sockets.  See the attached
patch, I need to test it a bit more before committing.

Thanks,

Paolo

* looking for [hidden email]--2004b/smalltalk--devo--2.2--patch-168 to compare with
* comparing to [hidden email]--2004b/smalltalk--devo--2.2--patch-168
M  kernel/PosStream.st
M  kernel/Stream.st
M  libgst/events.c
M  tcp/Sockets.st

* modified files

--- orig/kernel/PosStream.st
+++ mod/kernel/PosStream.st
@@ -245,13 +245,6 @@ collection
     ^collection
 !
 
-pastEnd
-    "The end of the stream has been reached.  Signal a Notification"
-
-    SystemExceptions.EndOfStream signalOn: self.
-    ^nil
-!
-
 status: aString
     "When working with a stream on strings, this method can be useful!
      Format of the output:


--- orig/kernel/Stream.st
+++ mod/kernel/Stream.st
@@ -300,12 +300,19 @@ tab
 ! !
 
 
-!Stream methodsFor: 'providing consistent protocols'!
+!Stream methodsFor: 'polymorphism'!
 
 close
     "Do nothing. This is provided for consistency with file streams"
 !
 
+pastEnd
+    "The end of the stream has been reached.  Signal a Notification."
+
+    SystemExceptions.EndOfStream signalOn: self.
+    ^nil
+!
+
 flush
     "Do nothing. This is provided for consistency with file streams"
 ! !


--- orig/libgst/events.c
+++ mod/libgst/events.c
@@ -190,9 +190,6 @@ file_polling_handler (int sig)
  }
       while (n == -1 && errno == EINTR);
 
-      if (n == 0)
- return;
-
       num_used_pollfds = 0;
       for (node = head, pprev = &head; node; node = *pprev)
  {


--- orig/tcp/Sockets.st
+++ mod/tcp/Sockets.st
@@ -818,6 +818,7 @@ available
 bufferContents
     "Answer the current contents of the read buffer"
     | result |
+    readBuffer isNil ifTrue: [ ^self pastEnd ].
     result := self readBuffer bufferContents.
     lookahead isNil ifFalse: [
  result := lookahead asString, result.




_______________________________________________
help-smalltalk mailing list
[hidden email]
http://lists.gnu.org/mailman/listinfo/help-smalltalk

On Tue, Oct 31, 2006 at 01:31:04PM +0100, Paolo Bonzini wrote:
But this doesn't fix my test case, should it?
I tried applying the patch to the latest cvs, and installing it and
tried it, but i got the same results: endless loop.

Robin


_______________________________________________
help-smalltalk mailing list
[hidden email]
http://lists.gnu.org/mailman/listinfo/help-smalltalk

> But this doesn't fix my test case, should it?
> I tried applying the patch to the latest cvs, and installing it and
> tried it, but i got the same results: endless loop.
>  
Yes, but this time I can see that 'End!' is reached.

Paolo


_______________________________________________
help-smalltalk mailing list
[hidden email]
http://lists.gnu.org/mailman/listinfo/help-smalltalk