Contribution: Mark Brown's bisection scripts

[bhcopeland]

Mark to share his ARM bisection setup.

Runs parallel builds across a k8s cluster and shares builds between
multiple bisections. Significantly faster than sequential.

Drop scripts into contrib/arm/ with a README covering dependencies
and how to run it.

[broonie]

These are mine, not Arm's. Arm only uses KernelCI.

[bhcopeland]

@broonie understood, my wording was incorrect here, updated the issue title. Let me know if you want another one for Arm's otherwise I'll leave it as is.

[broonie]

@broonie understood, my wording was incorrect here, updated the issue title. Let me know if you want another one for Arm's otherwise I'll leave it as is.

Huh, weird - I'd done that too.

[broonie]

Can't publish the code yet but algorithm is very simple:

Framework:

• Have a fixed set of test jobs, each of which runs a fixed set of tests
• Have a store of builds and a store of commit/test job results

Setting up bisections:

• Do a bare checkout from a central repo (so minimal data in the directory)
• Set up good/bad commits
• Search database for all commits in the bisect range (git bisect visualize) where this test job ran, feed in any results with 'git bisect good/bad.

Then iterate through all the bisections that have been set up running a script:

• Is there a build for the current bisect head? No, kick one off.
• Did the build pass? No, do a git bisect skip.
• Is there a test job for the current bisect head? No, kick one off.
• Did the job finish? Yes, read the test result.
• If result is missing or otherwise inconclusive skip the commit.
• If result is pass/fail then git bisect good/bad
• If bisect has only one commit left then report the result (email myself the bisect log)

All intermediate builds and test results from bisections also end up in the database for future use.