8259679: GitHub actions should use MSVC 14.28

[shipilev]

Current Windows GH Actions fails with:

configure: using /cygdrive/c/progra~2/micros~1/2019/Enterprise/vc/auxiliary/build/vcvarsx86_amd64.bat
configure: Setting extracted environment variables for x86_64
checking that Visual Studio variables have been correctly extracted... ok
checking for cl... [not found]
configure: error: Could not find a C compiler.
configure exiting with result code 1

I managed to figure out that MSVC build 14.27 is not really available. 14.28 is available instead. Note that Windows AArch64 build block already uses 14.28 in the same manner, so this change not only fixes the GH actions, it also keeps both blocks consistent.

---

Progress

• Change must not contain extraneous whitespace
• Commit message must refer to an issue
• Change must be properly reviewed

Issue

• JDK-8259679: GitHub actions should use MSVC 14.28

Reviewers

• Magnus Ihse Bursie (@magicus - Reviewer)
• Claes Redestad (@cl4es - Reviewer)

Download

$ git fetch https://git.openjdk.java.net/jdk pull/2126/head:pull/2126
$ git checkout pull/2126

[bridgekeeper]

👋 Welcome back shade! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

[openjdk]

@shipilev The following label will be automatically applied to this pull request:

• build

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

[mlbridge]

Webrevs

• 00: Full (7d4d91c)

[magicus]

LGTM.

[openjdk]

@shipilev This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8259679: GitHub actions should use MSVC 14.28

Reviewed-by: ihse, redestad

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been no new commits pushed to the master branch. If another commit should be pushed before you perform the /integrate command, your PR will be automatically rebased. If you prefer to avoid any potential automatic rebasing, please check the documentation for the /integrate command for further details.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

[shipilev]

/integrate

[openjdk]

@shipilev Since your change was applied there have been 2 commits pushed to the master branch:

• 061ffc4: 8249245: assert(((((JfrTraceIdBits::load(klass)) & ((JfrTraceIdEpoch::this_epoch_method_and_class_bits()))) != 0))) failed: invariant
• e60c992: 8259849: Shenandoah: Rename store-val to IU-barrier

Your commit was automatically rebased without conflicts.

Pushed as commit 6b4732f.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.

[mlbridge]

Mailing list message from David Holmes on build-dev:

Hi Aleksey,

On 18/01/2021 11:07 pm, Aleksey Shipilev wrote:

Current Windows GH Actions fails with:

configure: using /cygdrive/c/progra~2/micros~1/2019/Enterprise/vc/auxiliary/build/vcvarsx86_amd64.bat
configure: Setting extracted environment variables for x86_64
checking that Visual Studio variables have been correctly extracted... ok
checking for cl... [not found]
configure: error: Could not find a C compiler.
configure exiting with result code 1

I managed to figure out that MSVC build 14.27 is not really available. 14.28 is available instead. Note that Windows AArch64 build block already uses 14.28 in the same manner, so this change not only fixes the GH actions, it also keeps both blocks consistent.

Hard-wiring build numbers seems a maintenance PITA. :( Is there not a
way to do this that is independent of the actual build installed?

Would it not also be good to add a check for the existence of the build
that we do hard-wire? That way the problem will be very clear next time.

Cheers,
David

[mlbridge]

Mailing list message from Magnus Ihse Bursie on build-dev:

On 2021-01-19 00:13, David Holmes wrote:

Hard-wiring build numbers seems a maintenance PITA. :( Is there not a
way to do this that is independent of the actual build installed?

I agree that hard-wiring numbers in the code is bad. However, having
explicit version numbers for our dependencies is good. That is the
philosophy behind jib, and it has served us well. We should look at
finding a way to store the actual versioon numbers outside the
submit.yml file, though.

But what do you mean by "independent of the actual build installed"?

Would it not also be good to add a check for the existence of the
build that we do hard-wire? That way the problem will be very clear
next time.

It would be fantastic! Please tell me how to to that in the limited,
brain-dead Github Actions environment. :-(

/Magnus

Cheers,
David

[mlbridge]

Mailing list message from David Holmes on build-dev:

On 19/01/2021 11:48 pm, Magnus Ihse Bursie wrote:

On 2021-01-19 00:13, David Holmes wrote:

Hard-wiring build numbers seems a maintenance PITA. :( Is there not a
way to do this that is independent of the actual build installed?
I agree that hard-wiring numbers in the code is bad. However, having
explicit version numbers for our dependencies is good. That is the
philosophy behind jib, and it has served us well. We should look at
finding a way to store the actual versioon numbers outside the
submit.yml file, though.

I think this is a bit different to the way we use jib to manage our own
build environment.

But what do you mean by "independent of the actual build installed"?

We have AFAIK no control over what toolkit versions are installed for
Github actions, so they could change at any time. We update today to
request 14.28 and tomorrow they replace 14.28 with 14.29 and our builds
fail again even though they would likely be perfectly fine on 14.29. To
me if we just were able to say "Use latest VS 14.x" then that would
suffice most of the time. Occasionally we could hit an issue where our
sources are not yet compatible with the latest VS but hopefully that is
less frequent than the frequency these VS builds change.

Would it not also be good to add a check for the existence of the
build that we do hard-wire? That way the problem will be very clear
next time.
It would be fantastic! Please tell me how to to that in the limited,
brain-dead Github Actions environment. :-(

If we put those hard-wired version/build numbers into a file readable by
configure, then couldn't configure check it?

Cheers,
David

/Magnus

Cheers,
David

[mlbridge]

Mailing list message from Magnus Ihse Bursie on build-dev:

On 2021-01-20 06:48, David Holmes wrote:

On 19/01/2021 11:48 pm, Magnus Ihse Bursie wrote:

On 2021-01-19 00:13, David Holmes wrote:

Hard-wiring build numbers seems a maintenance PITA. :( Is there not
a way to do this that is independent of the actual build installed?
I agree that hard-wiring numbers in the code is bad. However, having
explicit version numbers for our dependencies is good. That is the
philosophy behind jib, and it has served us well. We should look at
finding a way to store the actual versioon numbers outside the
submit.yml file, though.

I think this is a bit different to the way we use jib to manage our
own build environment.

But what do you mean by "independent of the actual build installed"?

We have AFAIK no control over what toolkit versions are installed for
Github actions, so they could change at any time. We update today to
request 14.28 and tomorrow they replace 14.28 with 14.29 and our
builds fail again even though they would likely be perfectly fine on
14.29. To me if we just were able to say "Use latest VS 14.x" then
that would suffice most of the time. Occasionally we could hit an
issue where our sources are not yet compatible with the latest VS but
hopefully that is less frequent than the frequency these VS builds
change.

I think that depends on how we view the GHA testing, which still is not
clear. Or rather, each OpenJDK developer seem to have their own opinion
on this, and no official guidelines to help us.

I think we need to make the most effort we can to control the GHA
environment. That means requesting a specific version of VS, and having
a fail-fast if that is not available. That way, we get to know when GHA
environment changes.

Which is how things work today.

Would it not also be good to add a check for the existence of the
build that we do hard-wire? That way the problem will be very clear
next time.
It would be fantastic! Please tell me how to to that in the limited,
brain-dead Github Actions environment. :-(

If we put those hard-wired version/build numbers into a file readable
by configure, then couldn't configure check it?

So just to be clear: configure *did* check for the existence of VS, and
it did not find it, and thus it aborted and complained. So we did have
this fail-fast functionality. But since the GHA environment contained a
half-broken, half-working installation of the VS version in question,
the error message from configure was not really clear.

This PR was about updating the version of VS required by the GHA build
script. Unfortunately, I think it's not possible to either:

1) move the version number to a separate configuration file, instead of
hard-coding it in the yaml source

2) have the submit.yml GHA script check for the availability of the
specific version of VS before launching configure.

but it might be just my limited knowledge about the system. If it were
possible, I think we should do both.

/Magnus

Cheers,
David

/Magnus

Cheers,
David