Experimental Physics and
Industrial Control System

No, I don’t think that will fix it, because you have not increased NORD.  Any clients that use the new EPICS feature of subscribing to callbacks with a length of zero only receive NORD elements in the callback.  So the string they receive won’t be Nil terminated, even though the waveform record itself is now Nil terminated.

 

But you can’t increase NORD in devAsynOctet, because that device support must also be able to handle binary byte arrays, and increasing NORD would mess that up.

 

Mark

 

 

From: Eric Norum [mailto:wenorum@lbl.gov] 
Sent: Monday, November 12, 2012 2:58 PM
To: Mark Rivers
Cc: [email protected] talk
Subject: Re: strange EPICS glitch

 

Would this fix things?

 

--- devAsynOctet.c     2012-06-25 15:20:07.000000000 -0700

+++ devAsynOctet.c.proposed          2012-11-12 12:55:29.000000000 -0800

@@ -657,17 +657,21 @@

 static void callbackWfRead(asynUser *pasynUser)

 {

     devPvt         *pdevPvt = (devPvt *)pasynUser->userPvt;

     waveformRecord *pwf = (waveformRecord *)pdevPvt->precord;

     size_t         nBytesRead;

     asynStatus     status;

 

     status = readIt(pasynUser,pwf->bptr,pwf->nelm,&nBytesRead);

-    if(status==asynSuccess) pwf->nord = nBytesRead;

+    if(status==asynSuccess) {

+        pwf->nord = nBytesRead;

+        if (pwf->nelm > pwf->nord)

+            *((char *)pwf->bptr + pwf->nord) = '\0';

+    }

     finish((dbCommon *)pwf);

 }

 

 static long initWfWrite(waveformRecord *pwf)

 {

     asynStatus status;

     devPvt     *pdevPvt;

 

 

On Nov 12, 2012, at 12:43 PM, Mark Rivers <[email protected]> wrote:

Hi Andrew,

I will change asynPortDriver so that the setStringParam() method adds a Nil and passes that length to device support on readOctet or asynOctet callbacks.  That will fix the problem with areaDetector, which is probably the main driver class using long strings for path names, etc.

This cannot be moved up to asynWfOctetRead/asynWfOctetWrite device support in asyn/devEpics/devAsynOctet.c, because the asynOctet support for the waveform record needs to be able to handle binary data.  However, we could add a new device support in there that was specifically for strings, e.g. asynWfOctetStringRead/asynWfOctetStringWrite.  These could enforce the Nils in both directions. 

 

-- 
Eric Norum
[email protected]