Joe and Zach--

As we get closer to shipping a Beehive release, it would be great if we could get this controls test framework issue fixed. Still seems right to fail the Ant build when the TCH tests fail. This will still work with Cruisecontrol if you set a flag "cc.mode" when running in that environment. Would it be possible to address this in the next few weeks?

Thoughts?

Eddie



Eddie O'Neil wrote:

Definitely understand the cmd.exe issues. :)

But, isn't this just a matter of setting a flag for whether to fail a test run? It would be an addition to what we are doing now.

  I'd propose that this work as follows:

- default TCH behavior is to fail the build in the case of a test failure / abort / etc. TCH does this by setting an Ant property, for example "test.failure" that can be used in Ant like:

<target name="drt.errors" if="test.failure">
    <echo>!!!!! Errors or failures occurred running tests !!!!!</echo>
    <fail message="Controls TCH Tests Failed"/>
</target>

- this behavior can be over-ridden in CruiseControl mode. The override will still print some failure message (or even set a property) but doesn't set the "test.failure" property so that the build fails.

The latter is the behavior that we have today; the former is just an addition to the TCH behavior / API. This would make TCH more like JUnit in the way that failures can be controlled through properties.

  Thoughts?

Eddie




James Song wrote:

First, this is a windows only issue. Ant does not have this problem on
other platform. There is no need to do String searching for failure
detection on Linux/Unix.

Second, on Windows, someone mentioned python and perl works better than
cmd.exe. Maybe cmd.exe could be switched to python or perl?

-James

-----Original Message-----
From: Eddie O'Neil Sent: Wednesday, March 16, 2005 1:27 PM
To: Beehive Developers
Cc: Joe Pemberton; Zachary Smith
Subject: Re: Beehive Check-in Test (Linux) Failed


Hm. I've seen CC machine configs that can handle this sort of thing,

so we should dig into how to fix it.

IMHO, it's pretty important to be very obvious about test failures when running in a shell, so we should look at addressing this.

   Any other thoughts?

Eddie



James Song wrote:

I come across this issue when I set up a cc machine.

http://issues.apache.org/bugzilla/show_bug.cgi?id=30216

CC code boils down to the following:

<exec dir="." executable="cmd" failonerror="false"
           resultproperty="build.result">

The resultproperty does not return the correct value on Windows
sometimes.

That is why I switched the failure detecting code to look for "Build
Failed" from the log file.


-James


-----Original Message----- From: Joe Pemberton Sent: Wednesday, March 16, 2005 1:03 PM To: 'Beehive Developers'; Zachary Smith; James Song Subject: RE: Beehive Check-in Test (Linux) Failed

From what I understand, the reason the test run does not fail early

there is that for a CruiseControl run, it is useful to do the full


test

run (and not fail early if the controls tests have errors).  After the
full controls test build/run, CC greps for "BUILD FAILED", and fails


the

build accordingly.

James may know more about this..

-Joe



-----Original Message-----
From: Eddie O'Neil Sent: Wednesday, March 16, 2005 12:58 PM
To: Beehive Developers
Subject: Re: Beehive Check-in Test (Linux) Failed


My take is that those three test aborts should have failed the


entire

Ant process at the point where the controls tests failed.

Without that, the tests fail and the build continues, and the Ant process can still end in BUILD SUCCESSFUL even if there were aborts.

  Probably good to follow the JUnit / Ant model here -- fail early.

  :)

Eddie



Joe Pemberton wrote:


Eddie,
I'm not sure I follow you.  I see the three aborts in the TCH run,


and


I see a BUILD FAILED at the end of the run.  Also, the CC run itself
failed.

-Joe
-----Original Message-----
From: Eddie O'Neil Sent: Wednesday, March 16, 2005 10:45 AM
To: Beehive Developers
Subject: Re: Beehive Check-in Test (Linux) Failed


A second-order problem is that there are three tests in the TCH


test

suite that were aborted (and even stack traced), but these didn't fail



the controls test suite. Look at this link and scroll down to the TCH



output:

 http://beehive01.bea.com/downloads/20050316052618/general/build.out

Zach, can you take a look at this? We saw the same thing with some



controls test failures last week.

Eddie



Daryl Olander wrote:



So it appears that there is a break in the controls checkin tests


that

was caused by the same checkin from this morning.


http://svn.apache.org/viewcvs.cgi?rev=157742&view=rev

CC was broken at 5:26 this morning.  There were two sets of failures,
the WSM Drts failed in addition the  controls check in test failed

http://beehive01.bea.com/downloads/20050316052618/controls/tch.cout

There is only a single checkin here.


On Wed, 16 Mar 2005 10:33:09 -0800 (PST), [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:




View results here ->



http://beehive01.bea.com/cruisecontrol/buildresults?log=log2005031610210

0











Reply via email to