The modelmetadataprovider does significant reflection caching, which shows
up as a hotspot. It's only of our most commonly used services, so not only
do a TON of instances get created, they are very heavy-weight because of
the two ConcurrentDictionary instances they have.
This is a major refactor of how IBinderMetadata interacts with model
binders and value providers. We're doing this to support better
extensibility for metadata in ApiExplorer.
You'll notice a bunch of deleted code in DefaultApiDescriptionProvider
that maps metadata marker interfaces to a fixed list of Api sources. This
is replaced now with IBindingSourceMetadata - which also replaces the
hierarchy of marker interfaces. Now user code can create an arbitrary
binding source and have a consistent API for model-binders,
value-providers and full-visibility in ApiExplorer as
well.
Additonally, there's some error checking in place that better enforces the
constraints we already have in the system. IE you can't create a 'greedy'
model binder that uses value-provider data.
Two additional enhancements are planned for followup PRs:
1. Add a BindingSource property to model-metadata. This will remove some
duplication, but I want to delay it because it would touch another 10 or
so files.
2. Add an extensibility interface for our 'special' model binders like the
file binder so these can show up in ApiExplorer as well.
expire files in razor file cache.
Add a functional test to ensure the compiler cache does not get
initialized until the first request to a View.
Fixes#1708
1. Creates a filter called FormatFilter. This will look at the format parameter if present
in the route data or query data and sets the content type in ObjectResult
2. It adds new options called FormatterOptions, that contains the map of format to content tyepe
3. A method in MVC options to add the formatter mapping
- #439 (2 of 3)
- correct namespaces
- correct `Resources` class and member names; add new resources
- add `RequestServices` property to `ClientModelValidationContext`
- adjust to modern `IUrlHelper` API
- add `IClientModelValidator` support in `DataAnnotationsModelValidator`
- move previously-unused `StringSplit()` to `RemoteAttribute` and rename
- rewrite `RemoteAttributeTest`
- improve test method names
Reduce number of `[Remote]` constructor overloads
- remove `AreaReference` enum and related `[Remote]` constructor overload
- use `null` or empty `string` as explicit reference to the root area
- generally reduce parameter validation; match `UrlHelper`
Cleanup
- correct Engineering Guidelines violations
- especially: add doc comments
- correct spelling error in `_additonalFieldsSplit`
nits:
- minimize `null` checks in `AdditionalFields`
- make `GetClientValidationRules` `virtual`; some subclasses use `new` today
- add tests of `DataAnnotationsModelValidator.GetClientValidationRules()`
- remove `builder.ToString()` calls since it appears
https://roslyn.codeplex.com/workitem/246 has been resolved or Moq has
worked around that issue.
- #964
- compute `ModelMetadata.Order` based on `[Display]` attribute
- property affects e.g. `@Html.DisplayFor()` generation for complex objects
- also affects order of messages in validation summaries
- test new scenarios involving `ModelMetadata.Order`
- per-property `ModelMetadata` and related tests
- validation and `HtmlHelper` tests
- add `HtmlHelperValidationSummaryTest` (which touches on #453)
- update ModelBinding functional test to show use of `[Display(Order = x)]`
nits:
- move more `NullDisplayText` bits into proper slots (just above `Order`)
- add doc comments for `ComputeNullDisplayText()`
- add more assertions in tests using `ModelStateDictionary.HasReachedMaxErrors`
- remove some trailing whitespace
- avoid `Assert.True()` & `Assert.False()`; split some assertions up
- `""` -> `string.Empty` in affected test classes
- rename "DefaultEditorTemplatesTest~~s~~" class and file to follow guidelines
- rename "ModelBindingTest~~s~~" class and file to follow guidelines
FYI #1888 covers a predictable (or even just stable) order in the UI
Changes here are all focused around MaxModelErrors on
ModelStateDictionary.
MaxAllowedErrors now defaults to 200 (same as options). This means that
constructing a new ModelStateDictionary with the default constructor will
use this default. There's a new constructor for creating a
MaxAllowedErrors with a non-default value.
The ControllerActionArgumentBinder is now responsible for setting the
value from options onto ActionContext.ModelState. This results in better
layering and guarantees the option is respected if someone uses
extensibility to call model binding.
ModelStateDictionary.CanAddErrors is renamed to MaxErrorsReached. We
wanted to change the behavior of this property, but realized that it's
very useful inside the model validation code, so opted to renamed.
There's also a bunch of doc cleanup inside ModelStateDictionary to
simplify things and improve clarity.
This change allows you to set global defaults for ApiExplorer on the
ApplicationModel. Additionally, we're more lenient about configuring
ApiExplorer = on with conventional routing. If you turn on ApiExplorer at
the application level, we'll just skip over all conventionally routed
controllers instead of throwing.
This is a cleanup PR to improve the common usage of
ModelMetadata.Properties.
We found placed in code where both .Count and the ability to index by
property name would be useful. I was able to cascade this and simplify the
ModelBindingContext as well.
- remove `explicitValue` variables from two `DefaultHtmlGenerator` methods
This follows up a [code review comment](https://github.com/aspnet/Mvc/pull/1834/files#r23120381)
related to `DefaultHtmlGenerator.GenerateCheckBox()`. Same issue appears in
`DefaultHtmlGenerator.GenerateRadioButton()`.
@pranavkm -
> Also a different variable name? There's a parameter isExplicitValue that
> follows which always has a false value. Reusing this name seems messy.
My response to that comment should not be repeated. @pranavkm was correct
that the only issue here is naming. `explicitValue` was too easily
confused with the `isExplicitValue` argument used a few lines later.
They had separate purposes: `explicitValue` related to the "checked"
attribute while `isExplicitValue` relates to the "value" attribute.
- Made @inject handle trailing semicolons identical to @using; essentially ignores it.
- Added parser, runtime/designtime codegen and functional tests.
- Added Microsoft.AspNet.Mvc.Common.Test.
- Transitioned pre-existing Microsoft.AspNet.Mvc.Common tests to the new test project.
- Updated transitioned tests to also work in CoreCLR (except ones with moq).
#1857
This change makes ApiDescription and ApiParameterDescription aware of all
of the new features we built into model binding for enhanced DTO support
(uber-binding).
The main change is that instead of sticking just to the declared
parameters on the action itself, we now traverse model metadata and break
the parameters down based on their logical data source.
This means that a model like the below will yield 3 parameters:
public class ProductChangeCommandDTO
{
public int Id { get; set; }
[FromBody]
public ProductDetails Changes { get; set; }
[FromQuery]
public string AdminComments { get; set; }
[FromServices]
public IProductRepository Repository { get; set; }
}
The 'Repository' will be hidden, as it's not related to user input.
Additionally, we treat different sources differently. In the
above example, 'Changes' is from the body and will be treated as a
leaf-node.
However if you use nested DTOs that are bound from the query string (using
[FromQuery]) or similar, we'll recursively explore to find as much
structure as possible.
This information is combined with data from the route template to give a
much more complete picture than we ever could in the past for parameters,
especially when DTO/Command pattern is used.
- add a couple more exclusions to .gitignore (recent VS additions)
- remove `<ProjectExtensions/>` elements
- update files that don't have the correct output directories
- remove dangling PrecompilationWebSite.kproj file
This change moves controller creation to the stage immediately before
model binding. The controller will be disposed/released before Resource
Filters run their 'OnResourceExecuted' method. Previously the controller's
lifetime surrounded all filter invocation.
Additionally, the Controller property is now gone from ActionContext, and
is moved to the 4 filter contexts that should have access to it
Action*Context and Result*Context.
This is a major change to how we handle the scenario where a controller is
a filter. We want to change the lifetime of the controller object, by
scoping it around action filters and result filters. This means that a
controller class can only implement action filters and result filters.
To implement #384 - we're creating a delegating filter class
'ControllerFilter' which will forward calls to the implementation of the
controller. This is discovered in the controller model and added to the
filter collection. This filter is removable as an opt-out of this feature.
The ControllerFilter only implements action filter and result filter, so
the new restriction about filter types on Controller is in place. A future
change will move the instantiation of the controller to after resource
filters.
This is a major change to how we handle the scenario where a controller is
a filter. We want to change the lifetime of the controller object, by
scoping it around action filters and result filters. This means that a
controller class can only implement action filters and result filters.
To implement #384 - we're creating a delegating filter class
'ControllerFilter' which will forward calls to the implementation of the
controller. This is discovered in the controller model and added to the
filter collection. This filter is removable as an opt-out of this feature.
The ControllerFilter only implements action filter and result filter, so
the new restriction about filter types on Controller is in place. A future
change will move the instantiation of the controller to after resource
filters.
This change adds support for our three-valued logic to the default value
handling part of the MutableObjectModelBinder.
The issue is that we want to look up a default value when a 'greedy' model
binder returns true but doesn't find a value.
We also don't want to call the property setter unless there is:
1). A value from model binding OR
2). A default value
In general all properties are get/set so filters can change them.
- some validate for not-null
- where we use services it's get/set also
Services are resolved in the Execute method if not provided.
A few more ActionResults that return a body have the ability to set a
status code now (optional).
- React to aspnet/Razor#221
- Modified existing TagHelpers to no longer rely on ContentBehavior and to instead utilize GetChildContentAsync, PreContent, Content and PostContent.
- #1685
- move `ValidationSummary` type to the `Microsoft.AspNet.Mvc` namespace
- update tests and samples to match
- remove tests for case-insensitivity of `ValidationSummary` property values
- part II of #1253
- an expected case in template .cshtml files
- expression has name `""`; led to `ArgumentException` in `ModelExpression`
- test `@Model` and `@model.Property` in unit and functional tests
- update baselines to match
nits:
- remove a few unecessary `@`s in .cshtml files
- correct field names & ids in ProductList.cshtml (`foreach` confuses MVC)
- led to correct valiation attributes as well
- part I of #1253 using new Razor capabilities
- update baselines to match latest code generation
- use new `CodeBuilderContext` and `TagHelperAttributeValueCodeRenderer` signatures
- test complex expressions in bound non-string attribute values
See #1695 for a detailed explanation. This change builds support into the
system for the case that a model binder returns true without setting a
value for the Model.
In this case, validation will be skipped if it's a top-level object.
Note that explicitly setting null will still run validation.
This is a demonstration of how to inject an IRouter in between traditional
routes and MVC's handler. This allows you to accomplish a variety of
things that were possible with WebAPIs handlers, but inside the routing
system.
The example here turns a header representing the user into a locale, which
is used to select a controller. You could do other things like reject the
route match or change link generation.
There is one subtle project change here, to allow the same to be possible
for attribute routing, we need to create the attribute route after running
the user's routing configuration code.
- Support for binding posted file to type IFormFile
- Support for multipart/form-data in FormValueProviderFactory
- Updated Mvc Sample
- Added relevant unit and functional tests
The action invoker no longer needs access to model metadata or to the
input formatter selector. This change removes the same as constructor
parameters and cleans up tests which use the invoker.
This is a new filter stage that surrounds the existing model binding,
action, and result parts of the pipeline. Resource Filters run after
Authorization Filters.
The other major change is to support one of the primary scenarios for
Resource Filters. We want a filter to be able to modify the inputs the
model binding (formatters, model binders, value providers, etc) -- this
means that those changes need to be held on a context object and preserved
so that they can be used in the controller.
So, IActionBindingContextProvider is removed - the ActionBindingContext
will be created by the invoker. For now it will be part of the action
context.
- StyleCop working again (handles C# 6.0 additions) though only locally for me
- disable some new rules:
- ConstFieldNamesMustBeginWithUpperCaseLetter
- InstanceReadonlyElementsMustAppearBeforeInstanceNonReadonlyElements
- StaticReadonlyElementsMustAppearBeforeStaticNonReadonlyElements
- StaticReadonlyFieldsMustBeginWithUpperCaseLetter
- PrefixCallsCorrectly
- correct remaining violations
- lots of long lines for example
- use more `var`; some manual updates since StyleCop doesn't check seemingly-unused blocks
nit: remove new trailing whitespace (was paranoid about adding it w/ fixes)
The ParameterModel and ParameterDescriptor have had a notion of
optionality for a while now, even though all parameters are treated as
'optional' in MVC.
This change removes these settings. Optionality for overloading in webapi
compat shim is reimplemented via a new binder metadata.
This change adds an interface for the functionality provide by
RouteConstraintAttribute, and adds support for configuration constraints
on actions/action-model.
Fix - When the model is passed in to a View, ViewDataDictionary sets it. During this process, we recurse through all the properties and create FastPropertyGetters for each of them. In this case, since it is an enumerable, the properties which we recurse through are not the elements of the collection but the properties of the Enumerable instead. i.e - Enumerable.Current. Creating getters for these properties are not necessary. The fix moves the property iteration step to a place where the properties are actually requested.
- Splitting TypeInformation class into two and separating their caches appropriately.
Absorbs the new IFileSystem interface. This change is to just address the breaking change introduced in IFileSystem.
Razor has to do the necessary changes to subscribe to the Watch event for expiring the modified files.
The issue here is that a model state error is added with the model name
'doubled'. This is on a fairly obscure code path and the code dates back
to the original WSR git checkin of webapi. There are no wsr tests that
verify this behavior.
The cause here is that our 'greedy' model binders (like
FromHeaderModelBinder) return 'true' whether or not they successfully
found a model, because they don't want other model binders to run.
This also has an effect on the validation system. That means that
validators will run and attempt to validate the model (which may be null).
That's that rare case where we get to this code path.
- update XML docs to reflect new HTML / custom attribute separation
- update `Exception` messages to use new attribute names
- update MVC tag helper sample to use new custom attribute names
- add missing `<input/>` tag helper `throw`s test
nits:
- reword a few comments and messages for clarity and consistency
- use `<exception/>` sections to describe what's thrown
- add "Reviewers" comments about current throws
- note `<a/>` and `<form/>` are slightly inconsistent with others: `throw`
if unable to override specified `href` or `action` attributes; rest
`throw` only if custom attributes are inconsistent e.g. `<input/>` with
`asp-format` but no `asp-for`
- create test tag helpers after all property values are available
1. Updated ViewComponent to exposes similar properties to the existing ones in controller where
appropiate. We've left out Resolver for being a bad pattern (just inject the dependency on the constructor
or use Context.RequestServices to access it if needed) and Response as although available through the Context
property, it shouldn't be used/modified in a ViewComponent.
2. Updated ViewViewComponentResult to follow a similar pattern as ViewResult where the constructor is
parameterless and elements like ViewEngine are resolved during execution if the user does not set the
associated property on the object.
3. Updated ExecuteAsync in JsonViewComponentResult to remove the unnecessary pragma and async keyword from the
signature and to use Task.FromResult(true) instead.
4. Cleaned up ViewViewComponentResult tests.
- #EngineeringDay
- license present but incorrect in just a few files
- skip generated files such as Resources.Designer.cs and files under
test\Microsoft.AspNet.Mvc.Razor.Host.Test\TestFiles\Output
- #EngineeringDay
- VS does not yet format auto-properties nicely; reverted what it did
Also revert changes under
- test/Microsoft.AspNet.Mvc.Razor.Host.Test/TestFiles
- #EngineeringDay
- Total replaced: 660 Matching files: 270 in *.cs
- Total replaced: 250 Matching files: 32 in all other files
- Total replaced: 22 Matching files: 8 in a few stragglers
Did not change files under following directories
- test\Microsoft.AspNet.Mvc.Razor.Host.Test\TestFiles\Output
- test\Microsoft.AspNet.Mvc.FunctionalTests\compiler\resources
- test\WebSites\TagHelpersWebSite
(Razor generates trailing whitespace in a case or two)
For each of these TODOs:
- If there's an active bug tracking the work, and the TODO provides
something of value, I left it and standardized the formatting. I also
added comments to the bug.
- If the comment provided no value (implement feature X when we do feature
X), I deleted it with impunity.
- If the comment was stale (won't fix or just out of date), then we
removed it uncerimoniously.
There was a single TODO that was actually actionable, so I enabled that
test.
- TagHelperAttributeDescriptors changed to be lighterweight and not depend on PropertyInfo, had to modify our use of them to work with the new contract.
- #1523
- remove `TagHelperOutput.Merge()` extension method entirely
- test tag name preservation with all MVC tag helpers
- `<input/>` tag helper generation of a checkbox wasn't previously tested
nits:
- fix argument order in a couple of `Assert.Equal()` calls
- remove use of "original tag name"
This change enables some compatibility scenarios with MVC 5 by expanding
the set of legal ways to configure attribute routing. Most promiently, the
following example is now legal:
[HttpPost]
[Route("Products")]
public void MyAction() { }
This will define a single action that accepts POST on route "Products".
See the comments in #1194 for a more detailed description of what changed
with more examples.
Rather than throwing here, this does what routing does. If request
services aren't set, we just create our own scope.
This will NOT create an extra scope if request services are already set.
- clean up "the the" in XML comments
- simplify refactoring VS did when I renamed `GetHtmlHelperForViewData()`
- fix existing issue in `HtmlHelperCheckboxTest.CheckBoxReplacesUnderscoresInHtmlAttributesWithDashes()`
(was using a `HtmlHelper<ViewDataDictionary<TestModel>>`)
This is the MVC companion to https://github.com/aspnet/Routing/pull/122
As routing flows, routes replace the route data and mutate a copy. This
allows users to make changes that dirty the data without affecting
undesired state changes.
We also add the 'next' router for diagnostic purposes.
- value may remain in the `FormContext` beyond `</select>` end tag but will
be cleaned up at the `</form>` end tag of the containing `<form/>` element
- `SelectTagHelper` called prior to helpers for contained `<option/>`s and
not again later
- adjust mock setups to handle new `GenerateSelect()` call
- add assertions for expected `FormContext.FormData` entry
nit: mention #1468 in a test comment
- use new `ModelMetadata.HtmlEncode` property in HTML helpers
- specifically in default HTML display and editor object templates (e.g.
`@Html.DisplayFor()`) when value is non-`null` and the template is invoked
with template depth greater than 1
- similar to MVC 5.2 commit [2b12791aee4f](https://aspnetwebstack.codeplex.com/SourceControl/changeset/2b12791aee4ffc56c7928b623bb45ee425813021)
nits:
- remove dupe `null` check in `DefaultDisplayTemplates.ObjectTemplate()`
- move backing fields initialized with constants together in `ModelMetadata`
The change here is to always use the provided formatter, instead of using
it as a fallback. This is much less surprising for users.
There are some other subtle changes here and cleanup of the tests, as well
as documentation additions.
The primary change is that we still want to run 'select' on a formatter
even if it's the only one. This allows us to choose a content type based
on the accept header.
In the case of a user-provided formatter, we'll try to honor the best
possible combination of Accept and specified ContentTypes (specified
ContentTypes win if there's a conflict). If nothing works, we'll still run
the user-provided formatter and let it decide what to do.
In the case of the default (formatters from options) we do conneg, and if
there's a conflict, fall back to a global (from services)
JsonOutputFormatter - we let it decide what to do.
This should leave us with a defined and tested behavior for all cases.
- `null` checks can generally be removed but in one case needed to
special-case metadata for `object`
nit: remove duplicate empty `string` check in
`ExpressionMetadataProvider.FromStringExpression()`
- ensure correct `ViewData.ModelState` value in a `RazorPage` instance
- let view components inherit `ViewData.Model` from surrounding context
- do not create an isolated `ViewDataDictionary<TModel>` in one HTML helper
- allows previous `ViewDataDictionary<TModel>` constructor to be `internal`
- also add comments for that constructor
- change two `ViewDataDictionary` copy constructor calls to avoid uselessly
copying `Model` and related information from source
- `ViewDataDictionary<TModel>` constructors now pass `typeof(TModel)` to base
`protected` constructors
- move type compatibility checks into base `ViewDataDictionary`
- remove `ViewDataDictionary<TModel>.ModelMetadata` override
- don't retrieve `ModelMetadata` twice in a single constructor invocation
- remove newly-unused `protected` properties and use `private` fields in copy
constructors
Address longstanding problems found (see #1466)
- avoid reusing `ModelMetadata` after `Model` value changes
- reset `ModelMetadata` backing field in `Model` setter
- `Model` and `ModelMetadata.Model` could previously get out of sync
- carry `ModelMetadata` forward from an outer scope only if `Model` matches
- previously two scopes could have different `Model` values but share their
`ModelMetadata` (and `ModelMetadata.Model`)
- related to previous item but didn't require `Model` setting; switching
to a property of the same type as containing `Model` was enough
- avoid NRE if `ViewDataDictionary<int>.Model` is read before it's written
- problem affected all non-`Nullable` value types
- `ViewDataDictionary.ModelMetadata` setter should throw if value is `null`
nits:
- add and reword doc and code comments
- `ViewDataDictionary<TModel>` constructors should only inherit base's
parameter descriptions; have more information in the derived class
- make a few `ViewDataDictionary` properties get-only
- clean up `using` statements in `ViewDataDictionary<TModel>`
- make two constructors `internal`
This change adds the concept of a full-name to viewcomponents. View
components can be invoked using either the short name or long name. If the
provided string contains a '.' character, then it will be compared against
full names, otherwise it will be matched against short names only.
The short name is used for view lookups.
If the name is explicitly set via ViewComponent attribute, then the full
name is the name provided. The short name is the portion of the name after
the last '.'. If there are no dots, then the short name and full name are
the same.
If the name is not set explicitly, then it is inferred from the Type and
namespace name. The short name is the Type name, minus the 'ViewComponent'
suffix (if present). The full name is the namespace of the defining class,
plus the short name.
Taking the suggestion here to move these to a sub-object. This is future
proof in the event that we need to capture more data for ApiExplorer, and
reads better.
ViewComponents and Controllers now follow the same rules exactly for what
types of classes they can be.
Also corrected a bug in a test for controllers. Closed-generic types can
be controllers, the test was wrong.
never that for `object`
- `ViewDataDictionary<TModel>.ModelMetadata` was for `object` after base
copy constructor got value from `ViewDataDictionary<object>`
- problem led to #1426 symptoms
- with copy constructor leaving `base.ModelMetadata==null` more often,
`ViewDataDictionary<TModel>.ModelMetadata` usually tracks `TModel` if
`Model==null`
nit:
- fix existing comment in main `ViewDataDictionary` copy constructor
- Cleaned up some existing bad code that worked around case sensitive TagHelperOutput.Attributes.
- Modified MergeAttributes to be case insensitive when it comes to merging class attributes.
- Added a test to verify the new MergeAttribute case insensitive class merging.
aspnet/Razor#186
- This involved adding the StringComparer.OrdinalIgnoreCase comparer to the TagBuilder's Attributes dictionary.
- Added tests to validate that all methods that made use of TagBuilder.Attributes abide by the new ignore case mechanic.
- Added two sets of tests to validate the new functionality of Object => Dictionary HTML helper tests.
- Modified a functional test that utilizes HTML Helpers to provide same attribute-different case objects.
- Fixed existing HTML helper tests to account for new ordering of attrbutes (dictionary no longer adds key value pairs, it sets them).
#1328
This also comes with a rename of the namespace
Microsoft.AspNet.Mvc.ApplicationModel to
Microsoft.AspNet.Mvc.ApplicationModels.
Also tuned up some parameter and variable names for increased
understandability.
Youngjune Hong
Ajay Bhargav Baaskaran
Ryan Nowak
Yishai Galatzer
Pranav K
N. Taylor Mullen
sornaks
Harsh Gupta
damianedwards
Mugdha Kulkarni
Kiran Challa
Doug Bunting
Anthony Sneed
Kirthi Krishnamraju
Kiran Challa
Chris Ross
Stephen Halter
Hao Kung
stanislavromanov
Troy Dai
Bruce Bowyer-Smyth
SonjaKhan
Saar Cohen
Praburaj
ianhong
Mugdha Kulkarni
Daniel Roth
lajones
Victor Hurdugaci
kanchanm
jacalvar
David Fowler