This is the mail archive of the cygwin-developers@sources.redhat.com mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: Serial blocking read


Nevermind.  I understand now.  I was confused by the fact that this code
should not have been triggered unless the 'ready_for_read' had indicated that
there was stuff to read.

Does the below patch work for you?  It avoids code duplication.

cgf

On Fri, Dec 15, 2000 at 09:58:42PM -0500, Christopher Faylor wrote:
>Ok.  Can you provide a little bit of an explanation on why this
>solves your problem.  It doesn't make any sense to me that it would
>by reading the code.
>
>cgf
>
>On Fri, Dec 15, 2000 at 06:53:31PM -0800, Dan Morris wrote:
>>My bad... I totally spaced on even giving a filename.  Here's a better diff :
>>
>>--- fhandler_serial.cc-orig     Fri Dec 15 18:45:26 2000
>>+++ fhandler_serial.cc  Fri Dec 15 00:28:04 2000
>>@@ -119,6 +119,13 @@ fhandler_serial::raw_read (void *ptr, si
>>            }
>>        }
>>
>>+      if (overlapped_armed) {
>>+          if (!ClearCommError (get_handle (), &ev, &st))
>>+           goto err;
>>+         else if (st.cbInQue)
>>+           inq = st.cbInQue;
>>+      }
>>+
>>       overlapped_armed = 0;
>>       ResetEvent (io_status.hEvent);
>>       if (inq > ulen)
>>
>>-Dan
>>
>>On Fri, 15 Dec 2000, Christopher Faylor wrote:
>>
>>> I think this is coming from fhandler_serial.cc but I don't know for
>>> sure.  Can you provide a unified diff?  A complete unified diff will
>>> show the filename and give a little more context.
>>> 
>>> The contributing link on http://cygwin.com should give you some pointers on how
>>> to do this and how to provide a patch.
>>> 
>>> cgf
>>> 
>>> On Fri, Dec 15, 2000 at 05:38:09PM -0800, Dan Morris wrote:
>>> >Has anyone else had problems with serial port reads blocking even when
>>> >O_NONBLOCK is passed to _read in the 1.1.6-1 DLL?
>>> >
>>> >I noticed that my reads were appropriately non-blocking most of the time, but
>>> >blocked whenever "overlapped_armed" was set, because in this case raw_read
>>> >never finds out how many bytes are in the serial port's buffer.  This patch
>>> >seems to fix the problem for me :
>>> >
>>> >121a122,128
>>> >>       if (overlapped_armed) {
>>> >>           if (!ClearCommError (get_handle (), &ev, &st))
>>> >> 	          goto err;
>>> >>           else if (st.cbInQue)
>>> >>             inq = st.cbInQue;
>>> >>       }
>>> >> 
>>> >
>>> >If anyone else has found a better solution to this problem, let me know...
>>> >otherwise perhaps this patch will fix things.

Sun Dec 17 19:03:52 2000  Christopher Faylor <cgf@cygnus.com>

	* fhandler_serial.cc (fhandler_serial::raw_read): Always find number of 
	bytes ready to be read whether overlapped_armed or not.

Index: fhandler_serial.cc
===================================================================
RCS file: /cvs/uberbaum/winsup/cygwin/fhandler_serial.cc,v
retrieving revision 1.9
diff -u -p -r1.9 fhandler_serial.cc
--- fhandler_serial.cc	2000/09/08 02:56:54	1.9
+++ fhandler_serial.cc	2000/12/18 00:07:05
@@ -79,15 +79,16 @@ fhandler_serial::raw_read (void *ptr, si
 			// if vmin > ulen then things won't work right.
 	  overlapped_armed = -1;
 	}
-      if (!overlapped_armed)
+
+      if (!ClearCommError (get_handle (), &ev, &st))
+	goto err;
+      else if (ev)
+	termios_printf ("error detected %x", ev);
+      else if (st.cbInQue)
+	inq = st.cbInQue;
+      else if (!overlapped_armed)
 	{
-	  if (!ClearCommError (get_handle (), &ev, &st))
-	    goto err;
-	  else if (ev)
-	    termios_printf ("error detected %x", ev);
-	  else if (st.cbInQue)
-	    inq = st.cbInQue;
-	  else if ((size_t)tot >= minchars)
+	  if ((size_t)tot >= minchars)
 	    break;
 	  else if (WaitCommEvent (get_handle (), &ev, &io_status))
 	    {

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]