This refactors the targets used to build the shared framework and its .zip files. There are lots of reasons motivating this: Arcade convergence, migration to VSTS, making it easier to build this locally, etc.
Changes:
* Moves move content of build/Sharedfx.{props/targets} into eng/targets/SharedFx.Common.{props/targets}
* Update the build to produce a `runtime.$rid.Microsoft.AspNetCore.App` package (not just the one with symbols in it)
* Refactor the targets which produce .tar.gz/.zip files into separate projects in `src/Installers/`
* Refactor installers, unit tests, and the framework projects to use ProjectReference to flow dependencies between different parts of the build.
* Makes it easier to build the shared framework locally (for the inner dev loop, you can run `dotnet build -p src/Framework/Microsoft.AspNetCore.App/src/ -r win-x64`)
As a part of converging repos, we no longer need this code. It is currently buggy and slow, and we can get away without by hard-coding the repo graph (which rarely changes)
Our policy since 1.0.0 has been to always cascade version updates in the packages we own. e.g. if Logging has a product change in 2.1.x, then Kestrel, EF Core, Mvc, etc also re-ship with the updated Logging dependency. This has been done for a variety of reasons:
* NuGet does not show updates for transitive dependencies, only direct ones
* NuGet does resolves the lowest compatible transitive dependencies
* ASP.NET Core ships to both .NET Framework (where transitive dependency version matters) and .NET Core (where it matters less if you use the shared framework)
While transitive dependencies is still an important scenario, this practice of always patching has led to bigger issues.
* High probability users will unintentionally upgrade out of the shared framework: #3307
* Conflicts with metapackages that attempt to use exact version constraints: aspnet/Universe#1180
* A quality perception issue: the high volume of new versions in servicing updates with only metadata changes has created the impression that new versions of packages may not be very important. It's also made it appear like there are more issues product than there really are.
* High volume of packages changing with only metadata changes. Of the last 301 packages published in a servicing update, only 11 contained actual changes to the implementation assemblies. (3.5%)
This change implements a system to verify a new, non-cascading versioning policy for servicing updates. This required changes to repos to pin version variables to that matter per-repo,
and to remove some of the restrictions and checks.
Incidentally, this should make defining new patches easier because it automatically determines which packages are or are not patching in the release.
List explicitly as .csproj files the scenarios for which the offline package cache is important
Produces new artifacts designed for various scenarios, such as:
* Docker (where xml doc files are not needed)
* Azure web apps (where 1.x SDKs must still be supported, but xml docs are not needed)
The PR #1175 was incomplete. This fixes the cascading effect of patching to 2.1.1
Changes:
- add CheckRepoGraph (ported directly from the release/2.0 branch)
- Update submodules
* Split compilation and tests into separate phases
* Ensure template tests are skipped, and reduce duplication between test/build repo targets
* Show summary at the end of which repos failed
- Remove unnecessary tasks and scripts
- Ensure the KOREBUILD_DOTNET_* environment variables are preserved in the docker build context
- Other MSBuild cleanup of the targets
- hosting tar.gz archives
- rpm installers for generic and red hat
- deb installers for debian.8, ubuntu 14.04, ubuntu 16.04, ubuntu 16.10
Create hosting targz archives
Nate McMaster
Nate McMaster
John Luo
=
Pranav K