[win] i-Builds tests do not finish since making Edge the default on windows due to workspace lock being held

[jukzi]

https://ci.eclipse.org/releng/job/AutomatedTests/job/ep435I-unit-win32-x86_64-java17/

[jukzi]

cc @BeckerWdf got time to help?

[BeckerWdf]

Sorry. From your description I don't know what to do.

[jukzi]

Sorry. From your description I don't know what to do.

Probably nobody does -> investigate what exactly went wrong, find out why, solve.

[BeckerWdf]

What does "DNF" mean?

[merks]

Did Not Finish.

[jukzi]

as reported in https://download.eclipse.org/eclipse/downloads/drops4/I20241211-0310/testResults.php

[jukzi]

SWT had 1 error then platform.ui DNF:

[HeikoKlare]

I just commented on that issue here: #2648 (comment)

It was very likely caused by merging this PR yesterday: eclipse-platform/eclipse.platform.swt#1637

Sorry that no one of us responded on the issues so far. I just pinged the colleagues to have a look. I am currently out of office and thus, unfortunately, cannot help except for pointing to the cause. At worst, we could revert that PR if there is no quick solution.

[fedejeanne]

I'm looking into it as we speak. The error log mentions the folder EBWebView so I will look closer to when the userdata directory for Edge was introduced in eclipse-platform/eclipse.platform.swt#1548.

Unable to delete file C:\Users\genie.releng\workspace\AutomatedTests\ep435I-unit-win32-x86_64-java17\workarea\I20241211-1800\eclipse-testing\test-eclipse\eclipse\update_configurator_folder\.metadata\.plugins\org.eclipse.swt\EBWebView\lockfile

@amartya4256 ping!

[sratz]

I think this is a timing issue.
The eclipse.platform.releng.aggregator\production\testScripts\configuration\sdk.tests\testScripts\test.xml script starts many runtimes in quick succession:

• Run tests
• Delete workspace
• Run tests
• Delete workspace
• ...

Now, with the Edge browser, we spawn msedgewebview2.exe child processes.
When shutting down the runtime, it looks like those processes linger around for a little while, holding on to a lock in the workspace.

What I am not quite sure about: Why does this happen after the org.eclipse.update.configurator.tests.AutomatedConfiguratorSuite but not after the other ones before it?

[jukzi]

When shutting down the runtime, it looks like those processes linger around for a little while, holding on to a lock in the workspace.

junit tests are responsible to wait until their resources have been cleaned. Especially a (the) failed JUnit test should not leak such resources. please adapt the code as necessary.

[sratz]

junit tests are responsible to wait until their resources have been cleaned. Especially a (the) failed JUnit test should not leak such resources. please adapt the code as necessary.

Not sure if the individual tests can (or should) influence this. They do not spawn those WebView2 processes explicitly.

If anything, there is some Browser instance that is not disposed correctly. I'll have a look at the org.eclipse.update.configurator.tests.AutomatedConfiguratorSuite test suite to see if I can find anything suspicious.

[fedejeanne]

I think this is a timing issue. The eclipse.platform.releng.aggregator\production\testScripts\configuration\sdk.tests\testScripts\test.xml script starts many runtimes in quick succession:

* Run tests

* Delete workspace

* Run tests

* Delete workspace

* ...

Now, with the Edge browser, we spawn msedgewebview2.exe child processes. When shutting down the runtime, it looks like those processes linger around for a little while, holding on to a lock in the workspace.

This was exactly what we thought and @amartya4256 is trying to validate this theory locally as we speak.

What I am not quite sure about: Why does this happen after the org.eclipse.update.configurator.tests.AutomatedConfiguratorSuite but not after the other ones before it?

Probably this configuration:
https://github.com/eclipse-platform/eclipse.platform/blob/master/update/org.eclipse.update.configurator.tests/test.xml#L37-L47

[fedejeanne]

@sratz I think I posted 2 seconds after you so I'll ping you gain, just in case you missed it :-)

What I am not quite sure about: Why does this happen after the org.eclipse.update.configurator.tests.AutomatedConfiguratorSuite but not after the other ones before it?

Probably this configuration: https://github.com/eclipse-platform/eclipse.platform/blob/master/update/org.eclipse.update.configurator.tests/test.xml#L37-L47

☝️