This change fixes call sites on the main request path for a simple site
(model binding, validation, views) that allocate boxed list enumerators.
Some cases aren't addressed by this change because the fix is too invasive
or requires changing an important contract to take IList instead of
IEnumerable. Will follow up on those case by case in order of importance.
- #2969
- add `ModelBindingMessages` for configuration and `IBindingMetadataProvider` overrides
- use `interface` to avoid `new` oddities when adding a setter to an `abstract` property
- add `IModelBindingMessages` to `ModelMetadata` for use in rest of the product code
- plumb the various bits through the system
- add integration tests using a custom `IBindingMetadataProvider`s to override messages
nits:
- remove unused resources
- use `AttemptedValue` and not `model` in `SimpleTypeModelBinder`
This change works around two mono issues that are blocking travis. I'd
like to have tests skip instead, but unfortunately that would mean not
running any validation tests on mono so this seems better.
- RuntimeHelpers.GetHashCode will sometimes crash the runtime
- PropertyHelper sometimes returns throws null-ref
We don't want these extensions methods defined in a wierd namespace, it's
straightforward and future-proof to just make these strongly-typed
collections.
- #2722
- make communication of errors from formatters to `BodyModelBinder` explicit
- `JsonInputFormatter` now adds errors to `ModelStateDictionary` with correct key
- change `InputFormatter.SelectCharacterEncoding()` to add an error and return `null` when it fails
- one less `Exception` case and removes some duplicate code
nits:
- improve some doc comments (more `<inheritdoc/>`, `<paramref/>` and `<see/>`)
- add another two `BodyValidationIntegrationTests` tests
Adds a new property, FieldName, to ModelBindingContext. The FieldName is
the name of whatever code-element is being bound, regardless of what
model-prefix is in use.
This is needed for cases like the Header model binder. We always want to
use the property/parameter name and we don't care about model prefixes.
- #2993
- use `ClosedGenericMatcher` to handle e.g. non-generic model types implementing requested interfaces
- reduce `IsAssignableFrom()` use since created binders use explicit casts i.e. handle explicit implementations
- also add more integration tests covering various collection key formats, some with validation errors
- merge a few `[Fact]`s into `[Theory]`s
- #2992
- use new properties to replace common helper methods
- still a few `Nullable.GetUnderlyingType()` calls
- creating `ModelMetadata` or sites lacking `ModelMetadata` access e.g. `ModelBindingHelper.ConvertTo()`
- #2836
Part 1: Use existing property values when recursing in `MutableObjectModelBinder`
- remove `ComplexModelDTO` because that indirection made fixing this issue more difficult and doesn't add value
- started with an old closed PR (#2241) which did some of this work
- correct `MutableObjectModelBinderTest` tests that exercised behaviour that can't occur
- the old `dto.Results` dictionary was never incomplete; nor could it contain `null` values
Part 2: Change `MutableObjectModelBinder` to pass complex property values into binding system
- model binding no longer trounces nodes in the model tree that aren't bound
- create model instances less often in `TryUpdateModel` scenarios
- refactor `EnsureModel()` to `GetModel()` and use appropriately
- reorder logic in `SetProperty()` and `AddToProperty()` to avoid copying to / from the same collection
- also cleans up some code duplication
nits:
- clean up `MutableObjectModelBinderTest`
- fix odd line wrappings and indentation
- use `nameof()` more
- use `string.Empty` more
- simplify a couple of `Returns()` expressions
- make assertions in `TryUpdateModelIntegrationTest` more readable
- no need to work through `ModelStateDictionary.Keys`
- also use `Length` instead of `Count()`; test code but we don't need Linq at all in that test class
some more tests.
This change reverts the behavior change from
a6ce9abab1 and adds more tests around the
scneario that was actually broken.
The right behavior is that unconvertable values result in a validation
error. There's no special behavior around value types and required values.
- #2633
- do not leave `ModelBindingResult.ValidationNode` as `null` when we hit the `null` `RawValue` special case
- move two bits of code together to make the special case more obvious
- add `ModelValidationNode` (that suppresses validation) when `HttpRequestMessageModelBinder` is successful
- also suppress validation of `HttpRequestMEssage` properties
- suppress validation in `CancellationTokenModelBinder`, `FormCollectionModelBinder`, `FormCollectionModelBinder`
- do not create a `ModelValidationNode` when validation fails in `TypeConverterModelBinder`
nits:
- improve some doc comments
- add a quick `HttpRequestMessageModelBinderTest`
- #2793
- add `ICollectionModelBinder`, allowing `GenericModelBinder` to call `CreateEmptyCollection()`
- adjust `CollectionModelBinder` and `DictionaryModelBinder` to activate model if default types are incompatible
- do not create default (empty) top-level collection in fallback case if Model already non-`null`
- change type checks in `GenericModelBinder` to align with `CollectionModelBinder` capabilities
- add special case for `IEnumerable<T>`
- correct `ModelMetadata` of a few tests that previously did not need that information
- #2907
- return `null` if `Filter()` finds no matching value provider in collection
- fix lack of defensiveness in model binders' use of `IBindingSourceValueProvider` implementations
- remove test workaround for unusual `CompositeValueProvider` behaviour
- #2705
- add `JQueryFormValueProvider` and `JQueryFormValueProviderFactory`
- carry some code forward from MVC 5; correct to follow current coding guidelines
- refactor `ReadableStringCollectionValueProviderTest` into abstract `EnumerableValueProviderTest`
- enables reuse in new `JQueryFormValueProviderTest`
- also run these tests in `CompositeValueProviderTest`
nits:
- do not create a duplicate `CompositeValueProvider` instance in `Filter()`
- correct garbled sentence in `IBindingSourceValueProvider` doc comments
- simplify `FormValueProviderFactoryTest` (no need for Moq) and correct test name
- correct test class / file names
- `CompositeValueProviderTests` -> `CompositeValueProviderTest`
- `FormValueProviderFactoryTests` -> `FormValueProviderFactoryTest`
The change here is that when we create the ModelValidationNodes to just
return them rather than adding them to the tree. For a very large model,
having these extra nodes in the tree would eventually cause an OOM.
We're going to be taking a more thorough look at this code separately,
hence the quick fix.
- #1418
- add new fallback binding in `DictionaryModelBinder`
- similar to MVC 5 approach but more explicit and with better key conversion support
- fix bugs in `PrefixContainer` encountered while adding new tests of #1418 scenarios
- did not handle entries like "[key]" or "prefix.key[index]" correctly
- refactor part of `GetKeyFromEmptyPrefix()` into `IndexOfDelimiter()`; share with `GetKeyFromNonEmptyPrefix()`
- extend `ReadableStringCollectionValueProviderTest` to cover bracketed key segments
nits:
- remove use of "foo", "bar", and "baz" in affected test classes
- `""` -> `string.Empty`
- `vpResult` -> `result`
- Previously `ModelBindingResult.IsModelSet` would be set to true if type conversions resulted in empty => `null` values for ValueTypes. This resulted in improper usage of `ModelBindingResult`s creating null ref exceptions in certain cases.
- Updated existing functional test to account for new behavior. Previously it was handling the null ref exception by stating that errors were simply invalid; now we can provide a more distinct error.
- Added unit test to validate `TypeConverterModelBinder` does what it's supposed to when conversions result in null values.
- Added additional integration tests for `TypeConverterModelBinder`.
#2720
- Removed TaskHelper and refactored with ClosedGenericMatcher
- Removed TypeHelper
- Moved custom encodings to InputFormatter
- Moved ObjectToDictionary to PropertyHelper
- Removed respective tests and test projects
- was trying out rules matching frequest PR comments (then)
- did a manual scan to find new instances of same issues
- "" -> `string.Empty`
- `String` -> `string` and similar
- fill empty XML doc elements
- ignored `JsonPatchDocument<TModel>`; just too many empty elements
- corrected missing / extra / out-of-order `<param>` descriptions
- `xml-docs-test` detects incorrect external references but not these local issues
- cleanup duplicate code now that #2445 is fixed
- update unit tests using old `ModelBindingContext` setups
- fix (just) one integration test where `MutableObjectModelBinder` incorrectly calculated
`isTopLevelObject` and returned a non-`null` model
- undo temporary changes in `BodyModelBinderTests` due to increased reliance on incorrect
`isTopLevelObject` in #2445 fix
nits:
- combine tests that are now duplicates
- beef up coverage of some `MutableObjectModelBinderTest` cases
- remove unused `using`s
- part II of II for #2445
- `FormCollectionModelBinder` is an exception because container is not user-provided
- no `ModelState` entry added
- enable tests that #2445 was blocking
- fix these and other tests expecting different `ModelState` entries
- simplify logic in `FormFileModelBinder`
`ValueProviderResult`
- remove `protected` setters and parameterless constructor
- no scenario for their use in subclasses; however `ConvertTo()` remains `virtual`
- add single-parameter constructor
- use in most of the greedy and type-matching model binders
- add doc comments throughout class
nits:
- use new `ValueProviderResult` constructor in many existing tests
- `""` -> `string.Empty` and `vpr` -> `valueProviderResult` in `ValueProviderResultTest`
- improve some test names in `BodyValidationIntegrationTests`
- do not check `Message` of a Json.NET `Exception`
- part I of II for #2445 (with a duplicate code PR to follow)
- needed for #2445 because new `ModelState` entries for values will make inconsisteny worse
- change `BodyModelBinder` to use same keys for all `ModelBindingResult`s and `ModelState` entries
- return fatal error result if formatter adds an error to `ModelState`
- update potential callers to avoid avoid ignoring `IsFatalError`
- fix test attempting to serialize all of `ModelState`
- will be borked with additional `RawValue`s in state
- fix two other tests that serialized `ModelState` but checked only `IsValid`
nits:
- address minor inconsistencies in `ModelBindingContext`
- use `System.Reflection.Extensions` package a bit more, where it's already referenced
- remove some unused resources
- #2664
- use new property to correctly determine `isTargetEnum` in `GetCurrentValues()`
- avoid `ArgumentNullException` in all cases where raw values are `enum` but target is not
- stop skipping tests blocked by #2664, exposing a couple more #1487 issues
- use new property instead of private `GetElementType()` methods where possible
- cleans up some duplicate code
- also remove redundant use of `IsCollectionType` and `ElementMetadata`
nits:
- move properties above methods in `ModelMetadata`
- avoid accidentally-incorrect "Remove Unnecessary Usings"
This is the first step is some more refactorings to come in the future
with the goal of making MVC less monolythic. This makes the core of MVC
more reusable and more in line with the design of other vNext platform
components.
With this change, Mvc.Core contains just the minimal guts needed to build
a working app.
- Action Discovery
- Action Invoker
- Filters
- ObjectResult
- Model Metadata
- Model Binding
- Formatters
- Validation System
And yes, we are aware of the irony of 'minimal MVC' not including the view
system. The idea is that this is the kernel of an MVC app, and anything
real is layered on top.
The most noticable impact of this change is that MvcOptions has been blown
apart into more managable chunks. See the various ConfigureMvc*** methods.
The new Mvc.Extensions package is a placeholder while we evaluate and tune
the new definitions. Expect more changes as features are move to their own
packages, and in some case their own repositories.
For now there is no experience to bootstrap an Mvc.Core app. That's coming
next.
Ryan Nowak
Pranav K
Stefán Jökull Sigurðarson
Doug Bunting
Ben Adams
Kiran Challa
Chris R
Mugdha Kulkarni
N. Taylor Mullen
Ajay Bhargav Baaskaran
Harsh Gupta