[svick]

Fixes #7344.

Root cause

ML.NET invoked the Arcade Helix MSBuild wrappers without CI mode. When a Helix failure was already matched by a Known Build Error, the wrapper still returned exit code 1, causing Azure Pipelines to add a second generic Cmd.exe exited with code '1' or Bash exited with code '1' error. That generic error was not covered by the KBE, so Build Analysis remained red.

Changes

• Invoke the Windows and Unix Helix wrappers with -ci / --ci. Arcade reports the underlying failure to Azure Pipelines but suppresses the redundant wrapper exit-code error.
• Enable Arcade CI mode for the CodeCoverage build and test invocations.
• Resolve NuGetPackageRoot and LightGBMVersion from evaluated MSBuild properties so CodeCoverage finds TorchSharp and LightGBM native assets in Arcade's repo-local CI package cache without duplicating its package-root logic.

Validation

• Build 1589488 reproduced the original behavior: the deterministic failure was matched by temporary KBE [Temporary KBE] Build Analysis CI mode experiment for #7344 #7705, but the generic wrapper exit-code error remained unmatched.
• Build 1590576 confirmed that Helix CI mode removes the generic Cmd.exe / Bash exited with code '1' error.
• Build 1595561 passed after the temporary experiment test was removed.
• CodeCoverage build 1597259 passed with CI mode and dynamically resolved TorchSharp and LightGBM native paths.
• Clean MachineLearning-CI rerun 1597536 passed against the final merge commit. The GitHub Build Analysis check remains associated with build 1597258, whose only failed leg hit an unrelated transient connection failure while downloading .NET runtime 8.0.16.

Remaining caveat

The legacy Helix Azure Pipelines reporter can still emit a separate structured Test <name> has failed error that is not necessarily matched by the KBE. This PR deliberately does not disable that reporter because doing so would change which system gates test failures. Adopting the standalone Helix Job Monitor will disable the legacy reporter and remove that remaining duplicate error path.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.81%. Comparing base (44ede55) to head (8e81265).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7704      +/-   ##
==========================================
- Coverage   70.81%   70.81%   -0.01%     
==========================================
  Files        1421     1421              
  Lines      280034   280034              
  Branches    27621    27621              
==========================================
- Hits       198304   198301       -3     
+ Misses      73568    73563       -5     
- Partials     8162     8170       +8     

Flag	Coverage Δ
Debug	70.81% <ø> (-0.01%)	⬇️
production	64.25% <ø> (+<0.01%)	⬆️
test	90.75% <ø> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 12 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

🟡 Changes recommended

Fix the Unix-incompatible script syntax and centralize the Torch native package versions.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Enables Arcade CI mode for Helix and code-coverage workflows while configuring native dependency paths.

Changes:

• Adds platform-specific CI flags to build, test, and Helix invocations.
• Resolves package-root and LightGBM paths dynamically for coverage.
• Configures native TorchSharp dependencies for coverage tests.

File summaries

File	Summary	Findings
build/ci/job-template.yml	Updates shared CI, Helix, and coverage commands.	Critical: Windows cmd.exe syntax is used in Unix Bash steps. Moderate: Torch native package versions remain hard-coded instead of using centralized properties.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟢 Approval recommended

The changes match the stated CI failure mode and validated package-cache requirements without unresolved issues.

0 open findings

2 resolved since last review

• On Unix agents this script step is executed by Bash, but for /f, %%p, and set "..." are…
• The two Torch native package versions are still duplicated here instead of using the centralized…

🧠 Review effort: Balanced

[matouskozak]

Should we enable the -ci flag for the official CI also? (build/vsts-ci.yml). I'm not sure it uses the job-template...

[svick]

@matouskozak Good point. I think they should, so I enabled them.

[Copilot]

🟢 Approval recommended

No blocking issues remain, and the reported CI validation supports the changes in active pipeline configurations.

0 open findings

🧠 Review effort: Balanced

[matouskozak]

Are these the only native libraries we should add? I think TensorFlow also has some dependency.

[svick]

TensorFlow does have a native dependency, but it doesn't need to be added here, the dependency can be correctly resolved already.

[matouskozak]

Do you know why the CodeCoverage run skipped the Tensorflow tests https://dev.azure.com/dnceng-public/public/_build/results?buildId=1642172&view=ms.vss-test-web.build-test-results-tab ?

[svick]

Actually, TensorFlow should be added here, so I did that.

There are other TensorFlow (and more) tests that are still skipped. I will open a separate PR once this one is merged.

[rosebyte]

LGTM, please take a look on Matous' comments before merging.

[Copilot]

🟢 Approval recommended

No blocking issues remain, and the reported CI validation supports the intended behavior.

No open issues 🎉

_{Review effort: Balanced}