This change introduces a service facade for creating the application
model, running conventions, validating the result, and flattening the
model.
This is used in the ControllerActionDescriptorProvider and provides the
existing functionality for now. The ControllerActionDescriptorProvider
will process the results and turn each 'flattened' model into a single
action descriptor.
The next change will introduce another consumer of this service, that
turns the 'flattened' model into an EndpointModel so that it can be
exposed via Endpoint Routing's convention system.
---
The main considerations here:
The flattening semantics of application model are pretty complicated :(
The validation that CADP does is actually pretty in depth and might be
really low value... Errors with writing route templates do happen, and
those will be caught by the routing system eventually.... Errors with
duplicate route names are similar... Errors with 'mixed' attribute and
conventional routing are not at all common. I don't think I've ever seen
an issue get filed about this. I did the work to port all of this stuff
forward but I'm not totally sure it's valuable - however, I don't really
want to make an argument for its removal. These are just some random
thoughts to keep in mind if you're reviewing this 👍
* React to Razor.Design package removal
* Remove references to packages (Razor.Design, Razor.Extensions) solely used to bring in compiler \ targets
* Target netcoreapp3.0 in samples and tests to allow Sdk to infer values
* Cleanup InferParameterBindingInfoConvention
* Infer BindingSource for collection parameters as Body. Fixes https://github.com/aspnet/Mvc/issues/8536
* Introduce a compat switch to keep 2.1.x LTS behavior for collection parameters
* Do not infer BinderModelName in InferParameterBindingInfoConvention
* Convert `RouteValueDictionary` values to `string` using `CultureInfo.InvariantCulture`
- #8578
- user may override this choice in one case:
- register a custom `IValueProviderFactory` to pass another `CultureInfo` into the `RouteValueProvider`
- values are used as programmatic tokens outside `RouteValueProvider`
nits:
- take VS suggestions in changed classes
- take VS suggestions in files I had open :)
This doesn't really accomplish our goals for 2.2 - I don't have a clear
scenario where I would tell a developer to use this VS something else.
Will reevaluate in 3.0
- should resolve issues with occasional strange MSBuild caching issues in this repo
- modeled after aspnet/Scaffolding#905
- follows aspnet/BuildTools#729 recommendation to check in global.config file
- see also Microsoft/msbuild#2914
- use newer KoreBuild
- `.\build.cmd -update /t:noop`
- #8180
- add an error when binding fails for top-level model
- same case as when MVC creates "default" / empty model i.e. `ParameterBinder` can't detect this
- update `CollectionModelBinder` subclasses and the various providers as well
- controlled by existing `MvcOptions.AllowValidatingTopLevelNodes` option
smaller issue:
- change `ModelBinding_MissingBindRequiredMember` resource to mention parameters too
- #7562 part 2
- add `OriginalModelName` to `ModelBindingContext`
nit: take VS suggestions, mostly to inline collection initialization in `FormFileModelBinderTest`
Currently MVC is still running the IActionConstraint implementations for
features that we've already moved into the routing layer. This has a
significant perf cost associated with, and so we want to skip it because
it's redundant. However if anyone has implemented their own
`IActionConstraint`-based features, they still need to just work.
This change takes the approach of skipping the action constraint phase
at runtime unless we see something 'unknown'. This is an all or nothing
choice, and will run action constraints if **any** action constraint we
don't special case exists. This is the most compatible behavior (running
redundant constraints) when the application is using constraints that
the developer implemented.
Another approach I considered was to eliminate these constraints as part
of the process of building ADs. I don't think that's ideal because
people have written code that introspects action constraints. We should
consider something like this in 3.0.