RFR: 8310183: Update GitHub Actions to use boot JDK for building jtreg

[Christian Stein]

On Fri, 16 Jun 2023 11:14:10 GMT, Jaikiran Pai <jpai at openjdk.org> wrote:

>> Please review this change to use the boot JDK for building jtreg when running on GitHub Actions.
>> 
>> This is a best-effort follow-up change to
>> - #14448
>> which didn't have the desired results - the `Bad address` error does still appear with using the pre-installed JDKs 11 and 17.
>> 
>> Tests using the boot JDK for building jtreg seem to be successful and more stable.
>
> I had a bit more look at this today. This most certainly is some issue in the `msys2` package that gets installed on these VMs. And this appears to be some recent regression. I went back and looked at some of my personal jobs which were passing without issues and I can see that those were on `7efe20baefed56359985e327d329042cde2434ff` commit of `msys2/setup-msys2 at v2` github action. It appears that there has been a recent release of this action around 3 weeks back and the GitHub actions jobs are now bringing in `cf96e00c0aab3788743aaf63b64146f0d383cee9` https://github.com/msys2/setup-msys2/commit/cf96e00c0aab3788743aaf63b64146f0d383cee9
> 
> The place where this fails when building jtreg, is when launching `javac` with a `CLASSPATH` environment variable set. There's no obvious issues in that variable value - a few commands before, similar values for that variable have worked fine with `javac`. This will need a bit more investigation to understand why `make` fails to launch `javac` with this version of `msys2`. 
> 
> For now though, I pinned the `msys2/setup-msys2` action in our JDK's github action's workflow file to the commit that was working and that got my builds past this issue. Without that change I can consistently reproduce the issue every time.
> 
> I have created a branch which that change and reverted the commit in this PR to see if the jobs run fine https://github.com/openjdk/jdk/pull/14516.  It's currently in progress https://github.com/jaikiran/jdk/actions/runs/5288890247. If that works, then perhaps we should just use that pinned version of `msys2/setup-msys2`, until we report and have this issue fixed in that package?

Thanks for the 🕵️ work, @jaikiran 

I would suggest to keep building jtreg with boot JDK and also to pin the version of `msys2/setup-msys2`. This way, we keep a bit more control over parts being in use by the sanity check workflow.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/14507#issuecomment-1594564746