[OpenJDK 2D-Dev] JDK-8013810: UnixPrintServiceLookup not returning consistent values

[Phil Race]

Patrick,
Maybe you need to be clear in the problem statement. I can't actually
find it anywhere in this email thread.  I'm reverse engineering to it as 
follows:

You got an IPPPrintService for lookup via

PrintServiceLookup.lookupPrintServices(null, null)


but a UnixPrintService via

attributes.add(new PrinterName(name, null));
PrintServiceLookup.lookupPrintServices(null, attributes)


If you read my email you should see that if you make the change
I proposed you will never enter the getNamedPrinterXXX() methods
if you are using CUPS, which seems like it should fix your problem
of different classes.

You don't really need to test the class and probably shouldn't.
service.equals(serviceByName) seems better.

And although getServiceByName() shouldn't enumerate all printers,
because that can be very costly in some configurations,
it likely should first check to see if we already enumerated all printers,
and then see if its in the list. As its written its really intended for the
case where no enumeration has been performed but if it already
happened there's no harm in leveraging that.

Put another, way does this patch (on its own, no other changes) fix your 
problem
It modifies the getPrinterByName() method. If not, why, not, and please 
print
out the actual class/service names seen by your test so we can see more 
clearly

diff --git a/src/solaris/classes/sun/print/UnixPrintServiceLookup.java 
b/src/solaris/classes/sun/print/UnixPrintServiceLookup.java
--- a/src/solaris/classes/sun/print/UnixPrintServiceLookup.java
+++ b/src/solaris/classes/sun/print/UnixPrintServiceLookup.java
@@ -366,6 +366,29 @@
          if (name == null || name.equals("") || !checkPrinterName(name)) {
              return null;
          }
+
+        if (printServices != null) {
+            for (int i=0; i<printServices.length; i++) {
+                if (name.equals(printServices[i].getName())) {
+                    return printServices[i];
+                }
+            }
+        }
+
+        if (CUPSPrinter.isCupsRunning()) {
+            try {
+                URL url = new URL("http://"+
+                                  CUPSPrinter.getServer()+":"+
+                                  CUPSPrinter.getPort()+"/"+
+                                  name);
+                printer = new IPPPrintService(name, url);
+            } catch (Exception e) {
+            }
+            if (printer != null) {
+                return printer;
+            }
+        }
+
          if (isMac() || isSysV()) {
              printer = getNamedPrinterNameSysV(name);
          } else {


BTW keep all revisions of your code around so that people can compare.
I have no reference to what the earlier version of your fix did.

-phil.

On 5/8/2013 12:48 AM, Patrick Reinhart wrote:
> Hi Phil,
>
> Sorry, I can not share your point of the methods 
> getNamedPrinterNameXX() always returning a UnixPrintService though. 
> That is exactly the point of inconsistency shown by the test case. As 
> far as I checked it the only referred places of those 
> getNamedPrinterNameXX() is exactly the getServiceByName() method 
> mentioned to place the fix.
>
> In my opinion those method names are misleading anyway. They both 
> called getNamedPrinterNameXX() even though they return a PrintService. 
> As I assume that they originally did return the active printer name 
> and the PrintService was created outside, as the usage of the 
> getAllPrinterNamesXX() methods - that are called when creating all 
> PrintService instances. You explained me before that in case of 
> hundreds of printers it's faster to just lookup the printer state of 
> one printer only - which is done within the getNamedPrinterNameXX()  
> methods.
>
> Patrick
>
>
> Am 08.05.13 01:38, schrieb Phil Race:
>> I am assuming this  current webrev replaced the previous
>> one:- http://reinharts.dyndns.org/webrev/
>>
>> getNamedPrinterNameSysV() and
>>
>> getNamedPrinterNameBSD()
>>
>> should not be changed as they should always
>> create a UnixPrintService and so can do this
>> directly. You are subverting the code in
>> there to first check for CUPS which shouldn't
>> be needed.
>>
>>
>> I haven' t tested this out but from inspection,
>> I think the problem is in getServiceByName() where
>> it is ignorant of the CUPS possibility.
>> It needs to have the isCUPSRunning() check and
>> if so create an IPPPrintService().
>>
>> So that's where you should place your call.
>>
>> That is likely the only change here that is really necessary.
>> The refactoring is OK but but not essential to the fix.
>>
>> -phil.
>>
>> On 5/7/2013 2:23 AM, Patrick Reinhart wrote:
>>> Hi Jennifer,
>>>
>>> I have changed the test to not use non internal packages now that 
>>> produces the same results for my case now.
>>>
>>> Now I still waiting for the feedback of Phil to get that fixed 
>>> correctly..
>>>
>>> Best regards
>>>
>>> Patirck
>>>
>>> Quoting Jennifer Godinez <jennifer.godinez at oracle.com>:
>>>
>>>> Yes I have and dicussed with Phil.  It looks pretty good but there 
>>>> may be a safer way to fix it since the fix is still using 
>>>> lpc/lpstat commands for CUPS. Also, the regression test should be 
>>>> modified to use non internal package. Phil will give his input on 
>>>> this too.
>>>>
>>>> Jennifer
>>>
>>
>