Use the Implicit SDK RID for RID-Specific Apps by Default

[nagilson]

Part 1 of 4 for #26503
Resolves #23539
Resolves #26028

The implicit RID is now inferred under the following options:

SelfContained=true,
PublishReadyToRun=true,
PublishSingleFile=true
PublishAot=true
PublishSelfContained=true.

PublishSelfContained can be added to make a project publish self-contained by default.

Removed code from before where RID is injected in the CLI and moved it into the MS Build target.
Some changes to how self-contained is inferred as well.

To observe changes:

The following worked before w/ Sarah's change but doesn't require extra computation in the CLI anymore:

•   dotnet publish --self-contained
  

• dotnet build --self-contained

The following would cause an error before:

•   Also observe the property if set in .csproj. 
  

• dotnet publish -p:PublishSingleFile=true

• dotnet publish -p:PublishReadyToRun=true

• dotnet publish -p:SelfContained=true

• dotnet publish -p:PublishAot=true

• dotnet build -p:SelfContained=true (Note this also works if set in csproj unlike before, like others)

The following did not mean anything before and now causes the correct behavior, validating it works it on publish scenarios through the CLI (and the CLI only ftr):

• Also check in the .csproj:
  - dotnet publish -p:PublishSelfContained=true

Broken by this PR Until Daniel's PR is merged into MSBuild fixing the bug which causes this to break:

Publishing a rid agnostic library.

What didn't change:

• dotnet publish -p:PublishTrimmed

TODO:

• Verify PublishReadyToRun and such actually use implicit RID, improve tests.
• Handle PublishAot
• Fix the NetSDK error in case something goes wrong
• verify publish self contained

[nagilson]

So Rid Agnostic projects are a whole other thing to consider. Working on that right now

[nagilson]

Note on that one failing test: SelfContained=false is not supposed to be used with PublishTrimmed and it becomes a bit circular. Already having a discussion about that one.

[richlander]

Can you document a set of CLI commands that would have had an error previously that now work because of this work? That would help a lot in reviewing your work. I'm not really the one to review all the code.

[nagilson]

Can you document a set of CLI commands that would have had an error previously that now work because of this work? That would help a lot in reviewing your work. I'm not really the one to review all the code.

EDIT 2: This info is stale now, check the top comment.

EDIT: Copied this to the top for better visibility, should make reviewing easier.
@richlander Note this was tested exclusively in windows environments.

The following worked before w/ Sarah's change but doesn't require extra computation in the CLI anymore:

•   dotnet publish --self-contained
  

• dotnet build --self-contained

The following would cause an error before:

•   Also observe the property if set in .csproj. 
  

• dotnet publish -p:PublishSingleFile=true

• dotnet publish -p:PublishReadyToRun=true

• dotnet publish -p:SelfContained=true

• dotnet publish -p:PublishAot=true

• dotnet build -p:SelfContained=true (Note this also works if set in csproj unlike before, like others)

The following did not mean anything before and now causes the correct behavior, validating it works it on publish scenarios through the CLI (and the CLI only ftr):

• Also check in the .csproj:
• dotnet publish -p:PublishSelfContained=true

Broken by this PR Until Daniel's PR is merged into MSBuild fixing the bug which causes this to break:

~~- Publishing a rid agnostic library. ~~

What didn't change:

• dotnet publish -p:PublishTrimmed

[nagilson]

Retargeting to 7.0.100.

[nagilson]

Marking as draft again as Daniel's change has flown into MSBuild and the SDK. Once I verify this change works with that and agnostic libraries aren't broken anymore, I will remark as open and ping.

[nagilson]

I am going to try to get the implicit RID in for RC2 by removing PublishSelfContained. For PublishSelfContained, I don't think is going to happen for RC2 as it looks like its going to require another change to msbuild to not break rid agnostic libraries. Removing it from this PR and putting it elsewhere once it's ready.

[nagilson]

@dotnet/domestic-cat May someone please take a quick look at this? Hoping to get it in for RC2.
Pinging domestic cat because I don't seem to be able to ping the sdk team

[joeloff]

I'm happy with the change.

My only concern would be to maybe reword some of the errors. Right now it reads as A runtime identifier for the property 'PublishAot' couldn't be inferred. and I wonder whether 'PublishAot' property reads better.

[nagilson]

Awesome, thanks @joeloff for your review. Remaining work is looking into one test which appears to fail flakily on helix.