Speaking of not being bullet proof, during testing of the fix to
wait for a specific prompt an intermittent failure was observed.
...
Sending command: trace methods 0x2a9
reply[0]: MyThread-0[1]
Sending command: cont
WARNING: message not recieved: MyThread-0[1]
Remaining debugger output follows:
reply[0]:>
reply[1]: Method exited: return value =<void value>,
"thread=MyThread-0", nsk.jdb.exclude.exclude001.MyThread.run(),
line=93 bci=14
reply[2]: 93 }
reply[3]:
reply[4]: MyThread-0[1]
# ERROR: Caught unexpected exception while executing the test:
nsk.share.Failure: Expected message not received during 420200
milliseconds:
...
The wait for message times out looking for "MyThread-0[1]".
A WARNING is printed and the "remaining debugger output"
shows that "MyThread-0[1]" is in the buffer.
I'm still investigating why the message match is not found.
Adding a final check before failing the wait for message
seems to workaround the problem.
diff --git a/test/hotspot/jtreg/vmTestbase/nsk/share/jdb/Jdb.java
b/test/hotspot/jtreg/vmTestbase/nsk/share/jdb/Jdb.java
--- a/test/hotspot/jtreg/vmTestbase/nsk/share/jdb/Jdb.java
+++ b/test/hotspot/jtreg/vmTestbase/nsk/share/jdb/Jdb.java
@@ -515,10 +515,11 @@
long delta = 200; // time in milliseconds to wait at
every iteration.
long total = 0; // total time has waited.
long max =
getLauncher().getJdbArgumentHandler().getWaitTime() * 60 * 1000;
// maximum time to wait.
+ int found = 0;
Object dummy = new Object();
while ((total += delta) <= max) {
- int found = 0;
+ found = 0;
// search for message
{
@@ -553,6 +554,12 @@
log.display("WARNING: message not recieved: " + message);
log.display("Remaining debugger output follows:");
receiveReply(startPos);
+
+ // One last chance
+ found = findMessage(startPos, message);
+ if (found > 0) {
+ return found;
+ }
throw new Failure("Expected message not received during
" + total + " milliseconds:"
+ "\n\t" + message);
}
On 9/20/18, 5:47 PM, Chris Plummer wrote:
Looks good. Still not bullet proof, but I'm not sure it's
possible to write tests like this in a way that will work no
matter what output is produced by the method enter/exit events.
Chris
On 9/20/18 10:59 AM, Gary Adams wrote:
The test failure has been identified due to the "int[2]"
being misrecognized as a compound prompt. This caused a cont
command to be sent prematurely.
The proposed fix waits for the correct prompt before
advancing to the next command.
Webrev: http://cr.openjdk.java.net/~gadams/8208473/webrev/
Issue: https://bugs.openjdk.java.net/browse/JDK-8208473
Testing is in progress.
