Add template for creating new experiments

[mrlacey]

Draft only - for tracking progress

For #4 (Template Infrastructure)

Still to do:

• Update to reflect changes in test projects
• Add doc(s) files
• Confirm what functionality (if any) should be in the generated classes
• Confirm what functionality (if any) should be in the generated samples
• Confirm what "working examples and helpful comments" are needed
• Need a naming convention for Sample pages (and the directory--or directories--they are in)
• Should the documentation file be visible inside the solution?
• Finally: double-check the generated solutions all work as expected

current issues:

• SourceGenerators issues when running the UWP sample head
• WASM sample head doesn't include the shell

[mrlacey]

How I'm (manually) testing this:

Generate a new project also called CanvasLayout but in a different folder. The diff the two "CanvasLayout" folders and verify the only differences are:

• Project Guids
• No functionality in the new src\CanvasLayout.cs file
• Only one sample page in samples\CanvasLayout.Sample\
• Different ports used in the WASM head
• Minor white space differences

Also

• Tests run in the newly generated version
• Sample apps all run in the newly generated version

Then, generate another new experiment but with a different name and compare it with the previously generated CanvasLayout version and confirm the only differences are:

• Project Guids
• Project names (there should be no results in a search for "CanvsaLayout" in the version with a different name.)
• Different ports used in the WASM head

Adding automated testing for this isn't going to be easy and is unlikely to be useful in the long term.

[michael-hawker]

Looking good, @Arlodotexe any feedback?

[michael-hawker]

Can we use the ProjectTemplateSamplePage here? That gets auto-replaced right? Would be easier to accidentally prevent multiple experiments having a SamplePage and leading to confusion later?

Or do we want to do like ProjectTemplateFirstSamplePage? And rename the namespace ProjectTemplate.Sample.FirstSample and folder to FirstSample to be consistent?

[mrlacey]

This would leave use with a lot of different experiments all creating files in a PT.S.FirstSample namespace which might look weird when they're put together ina a collective sample.

Do different sample pages have to be in separate directories? Do they need to be in directories at all--can't they just be in teh root of this project?

[michael-hawker]

I think we can just leave them in the root of the project without having folders for each. Or we don't have to create separate namespaces. The ToolkitSample attribute is going to expect that the id: nameof(SamplePage) is unique across all samples, so we don't want to have this be generically named and introduce the possibility of collisions.

[michael-hawker]

Since the template project should be its own stand-alone project. It should be buildable on its own still, eh?

If so, in this PR, we should swap out our 'experiment' job in the CI YAML to build the template solution here as well. That should help us ensure that we don't break things if we change the template or the props files, right?

Or maybe we can invoke the template command itself to create a dummy experiment in that job and try and build that? Maybe that'd be a better "integration" type test for the templates?

Thoughts?

[mrlacey]

Since the template project should be its own stand-alone project. It should be buildable on its own still, eh?

If so, in this PR, we should swap out our 'experiment' job in the CI YAML to build the template solution here as well. That should help us ensure that we don't break things if we change the template or the props files, right?

Good idea

Or maybe we can invoke the template command itself to create a dummy experiment in that job and try and build that? Maybe that'd be a better "integration" type test for the templates?

I'm not sure how (or if) this would be done but it's a nice idea.

[michael-hawker]

Or maybe we can invoke the template command itself to create a dummy experiment in that job and try and build that? Maybe that'd be a better "integration" type test for the templates?

I'm not sure how (or if) this would be done but it's a nice idea.

I'm thinking we can just add the two dotnet commands to install the template and invoke the template with some 'DummyTest' name within the existing yml file here:

Labs-Windows/.github/workflows/build.yml

Lines 200 to 203 in c7db198

	- name: MSBuild
	working-directory: ./labs/CanvasLayout
	run: msbuild.exe CanvasLayout.sln /restore -p:Configuration=Release

And then we'd just build it with that known path/name instead of CanvasLayout one here.

I think that should be pretty simple.

[mrlacey]

I have moved this out of draft as I think it's ready.
Currently failing due to changes introduced by @Arlodotexe which I assume are in preparation for some other changes being made. (I assume for #30 - but unconfirmed.)
Once those changes are in this should be good to go.

[Arlodotexe]

Hey @mrlacey, adding a second experiment uncovered an important issue with the sample and documentation registry systems and @michael-hawker asked me to investigate and fix. I'm getting things back in order - will be finished soon!

[mrlacey]

Hey @mrlacey, adding a second experiment uncovered an important issue with the sample and documentation registry systems and @michael-hawker asked me to investigate and fix. I'm getting things back in order - will be finished soon!

This is the issue I've been wanting there to me multiple experiments for.

[Arlodotexe]

All finished here, we should have no problems with multiple experiments now.

[Arlodotexe]

Rest of the code looks good from my end, just waiting on @michael-hawker to give the OK 🙂

[michael-hawker]

Looks good, I created #102 using this initially, so worked well. Will merge this and then test @Arlodotexe's fixes rebasing #102 and seeing how it works out.