Hi John,
Thanks for the reply, that was really helpful. The TestBarrier class
does the trick. That was the only class I needed, the class is quite
small, and I need to make some modifications to it, so I copied the
class into the databinding test bundle rather than create dependencies
to the platform tests.
Nigel
On 07/03/2014 15:40, John Arthorne wrote:
Hi Nigel, some thoughts:
- The most important thing is putting all blocking code into
jobs/threads and not run anything that may deadlock in the main test
thread because it can kill a whole test suite. All tests should be
forking off the real test, and then the main test waits with a bounded
timeout for completion.
- There is some basic infrastructure in the
org.eclipse.core.tests.harness bundle, such as TestJob and
TestBarrier. TestBarrier is used heavily in platform tests, but you
could use the Semaphore classes in java.util.concurrent with similar
effect.
- Essentially using a semaphore lets you strictly control the order
of execution between multiple threads, making the test predictable and
nearly equivalent to a single-threaded test. Tests need to be clearly
repeatable wherever possible, rather than just running a concurrent
activity 100 times and maybe it will fail, maybe not.
- Having tests rely on an unapproved library that everyone running
tests will need to install separately is a no-go. If it is really
valuable we can enter a CQ and get it added as a proper dependency.
Good luck, concurrency testing is really hard!
John
From: Nigel Westbury <[email protected]>
To: [email protected],
Date: 03/04/2014 07:52 AM
Subject: [e4-dev] Unit testing multi-threaded code
Sent by: [email protected]
------------------------------------------------------------------------
I was wondering if there were any recommendations on how we write unit
tests for multi-threading code. I am writing tests for
https://bugs.eclipse.org/bugs/show_bug.cgi?id=425704(making obserables
thread-safe).
I have used https://code.google.com/p/multithreadedtc/very effectively
on other projects. It is published under the BSD license, which we
probably can't commit, so would that mean users would have to install
manually to run the tests?
_______________________________________________
e4-dev mailing list
[email protected]
https://dev.eclipse.org/mailman/listinfo/e4-dev
_______________________________________________
e4-dev mailing list
[email protected]
https://dev.eclipse.org/mailman/listinfo/e4-dev
_______________________________________________
e4-dev mailing list
[email protected]
https://dev.eclipse.org/mailman/listinfo/e4-dev
