Refactor the `ci.psm1` to publish the code coverage and test packages.
Allow `CodeCoverage` configuration on non-windows.
## PR Context
We plan to run code coverage on Windows, Linux and macOS. These changes are needed to unblock those runs.
Add support in packaging.psm1 to produce a .msix AppX package. Update the docker image to use the new msix package type. Update the associated yml files so AzDevOps performs the build.
## PR Context
Enable publishing PSCore6 to Microsoft Store
Fixes#8919
Preserve user shortcuts pinned to Taskbar during MSI upgrade by not removing shortcuts in this case (assuming the user has not changed the installation directory), see https://stackoverflow.com/a/33402698/1810304
This also requires the Guid to not always be re-generated, which PR #7701 originally added to ensure shortcuts get removed when RTM and preview are installed, the underlying problem was rather that RTM and preview shared the same GUIDs, therefore the GUIDs are hard-coded again but different for RTM and preview, therefore the shortcuts will still always get removed on uninstall. But this also means those GUIDs should change when the default installation directory changes, i.e. in PowerShell 7. Should we write the code to already take this into account that it does not get forgotten?
Tested by first reproducing the issue by building installers locally (and bumping the patch version. Then the fix was applied to verify the solution, it. For this to take effect the version from which an MSI is being upgraded must have this fix already, i.e. if this fix got shipped in `6.2.1`, then on upgrading to it, the issue would still occur but when upgrading `6.2.1` to `6.2.2` the shortcut would start being preserved. I am wondering if we could maybe improve this to show effect earlier by trying to extract the used (auto-generated) GUIDs in the `6.2.0` and `6.2.0-rc` packages out and use them...
Please not that we probably need to take this out for `7.0` because the base installation directory will change. This also assumes that the user has not specified a different installation directory on upgrade but this is a bit of an edge case where I think other things might break as well.
Moved check if able to write to $PSHome as way to skip test to `BeforeAll` which already contained a check if running on Windows.
## PR Context
As part https://github.com/PowerShell/PowerShell/pull/9279, tests were updated to be skipped if the test requires writing to `$PSHome` but is not able to. However, these tests already had a skip mechanism in place so the additional check caused the test to run when it should have skipped.
Co-authored-by: Travis Plunk <github@ez13.net>
This PR optimizes the download tasks to reduce build times. After the change we see a order of magnitude reduction in download times for Windows package signing and also the compliance tasks.
In addition to that the PR also makes changes to organize the build artifacts in folders.
Removes appveyor stuff
## PR Context
<!-- Provide a little reasoning as to why this Pull Request helps and why you have opened it. -->
#8686 , appveyor removed
The `f` in Dockerfile is not uppercase, this breaks syntax highlighting for vscode on my client
<!-- Summarize your PR between here and the checklist. -->
## PR Context
`DockerFile` breaks syntax highlighting as mentioned and described above.
<!-- Provide a little reasoning as to why this Pull Request helps and why you have opened it. -->
The NuGet build task should wait on uploads for DEB job to complete.
## PR Context
There is a race condition when the build for DEB is completed but the upload has not. If the NuGet build job starts it may not get all the packages needed for building nuget packages.
Refactor macOS CI so that tests run in parallel
- Also:
- Fix bootstrap issue where curl was failing for a month (so obviously not needed)
- Change Windows test stages to use PowerShell Core (where we can) to avoid compat issues
## PR Context
This allows running all tests in about the same about of clock time.
I noticed a couple of inconsistencies when reading through the install bash scripts.
- Make documentation for switches consistent over all files.
- Replace all `sed` implementations of `lowercase` with a more maintainable `tr` implementation.
- Set the `OS` variable in every install script, making it so previously unused checks are actually used.
- Exit with a non-zero exit code when the script reaches an illegal state.
## PR Context
A lot of people, including myself, read the install scripts before executing them. While doing so I noticed they contained inconsistencies. For example, certain flags you can pass to the install script were either undocumented or had an incorrect description. This PR fixes some of these inconsistencies, which should make them easier to maintain and easier to read.
Co-authored-by: Travis Plunk <github@ez13.net>
* Add shim dll to redirect to appropriate runtime
* Build updates for creating global tool package
* Add building global tool to nuget.yml
* Add azure upload of nuget packages
Make test file result names unique
- Also, remove dead code that I found during investigation
Fixes#8978
## PR Context
Making the filenames unique, ensures that we can download the results from the artifacts and do our own processing of the results.
This removes some of the last mentions of the AppVeyor name. Don't change the title, we don't want the word to appear in the CLs
Also problem: the last references are in the changelogs but I don't want to modify them without committee approval.
## PR Context
it was removed, #8686
Split Windows CI into the following Jobs
1. Combine Build and xUnit, which published build artifact
1. Elevated CI tests
1. Unelevated CI test
1. Elevated everything else tests
1. Unelevated everything else tests
1. Package build and test
1. Verify xUnit result
## PR Context
Before this test:
- A CI run would take ~14 minutes
- A Feature run would take ~30 minutes
After this change, it always runs all tests and takes ~20 minutes.
Job `Unelevated everything else tests` is the long poll, taking ~4 minutes longer than the next shortest job. We should be able to get this back down to ~15 if we split that test job somehow.
