On 4/1/19 5:22 PM, Lance Andersen wrote:
A follow-up on this.
I ran this test 300+ times without failure on the internal Mach 5
machines including the one that it failed on (which was only 4 times
since January). This one system would run the test in approximately
6-7 minutes on average where as the the other window systems were
running running around 1 - 1.5 minutes on average.
After a side-bar conversation with Mandy/Alan it was suggested to
try to use the ToolProvider API vs ProcessBuilder where applicable
for invoking the javac and jar commands.
With the change, the system that was running 6-7 minutes was down to
28 seconds on average on the slow machine.
This is very good improvement.
The webrev can be found at:
http://cr.openjdk.java.net/~lancea/8216539/webrev.01/index.html
Looks okay to me.
Mandy
