Experimental Physics and
Industrial Control System

Dirk Zimoch <[email protected]> · Mon, 05 Feb 2007 14:45:31 +0100

Hi,
In StreamDevice, I do a pasynOctet->flush(pvtOctet, pasynUser) before doing any
writes to get rid of "old" replies. Asyn is not doing this automatically (and
also should not).
The flush() function works as if you do read() until no further data is
available and discards everything. But it might be significantly faster.
Dirk
Phillip Sorensen wrote:

Phillip Sorensen wrote:
I took some time to look at the asyn code this morning. I am looking at
ASYN 4-6.
Mark Rivers wrote:
I believe asyn is doing the flush operation that Emmanuel suggested.

As far as I can tell, there is no flush before writes at either the
devGpib level or the port driver level for drvAsynSerialPort or
drvAsynIPPort. What I did find was that in the drvAsynSerialPort.c file
that a timeout at this level causes a flush. There is no similiar flush
in the drvAsynIPPort.c code.
I suspect you are not getting out of order replies. I think you are
getting a timeout, but then stale data from some buffer is ending up
in the apparent data you received with the timeout.

The "out of order" data seems to be coming from the OS socket buffer (I
am running under linux). They are never flushed after a timeout.
This looks like a "possible" bug to me, and think that flushing the
buffer would do away with my "out of order error", BUT I don't think it
is the root cause of my problem.
After taking a second look at the code, I think I have found the root of
my problem. I was puzzled by the fact that the code seemed to work with
an older version of ASYN and that increasing the timeout parameter had
no effect on the problem.
It appears that when the timeout handling code was removed from the
drvAsynIPPort driver the SOCK_EINTR error return from the recv is now
treated as if it were a timeout. In the previous version of the code
that I have and in the drvAsynSerialPort code, the recv was inside a
for(;;) loop which would retry the read in the EINTR error state.
I will try to put together a patch to retry to the recv on SOCK_EINTR in
the next day or so.

Phil Sorensen
CHESS

--
Dr. Dirk Zimoch
Swiss Light Source
Computing and Controls
Paul Scherrer Institut
phone +41 56 310 5182
fax   +41 56 310 4413

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System