These changes are aimed at significantly improving the performance of
MVC/Razor when a large amount of content is in play or a large number of
TagHelpers are used.
A few issues addressed:
- Buffer sync writes after a flush has occurred so that we can write them
asyncronously. The issue is that an IHtmlContent can only do sync
writes. This is very bad for Kestrel in general. Doing these writes
async is better for our overall perf, and the buffer that we use for it
is from the pool.
- 'Flatten' ViewBuffers when possible. A page with lots of TagHelpers can
end up renting a ViewBuffer and only write 2-3 things into it. When a
ViewBuffer sees another ViewBuffer we can either steal its pages, or
copy data out and 'return' its pages. This lets us use 3-4 buffers for a
large Razor page instead of hundreds.
- #3215
- add new accessor properties to `IModelBindingMessageProvider` and plumb them through
- use in `ModelStateDictionary` when handling a `FormatException` or `OverflowException`
- use in `ValidationHelpers` when handling a `ModelError` with `null` `ErrorMessage`
- add new `ModelExplorer` parameter to `IHtmlGenerator.GenerateValidationMessage()`
- plumb through to `ValidationHelpers.GetModelErrorMessageOrDefault()`
Started from work @kichalla did on the `kiran/movemessages-to-messageprovider` branch in #3775.
nits:
- use helper methods more consistently in `HtmlHelper<T>`; slightly improves error checking
- remove unused `Resources` class from `Microsoft.AspNet.Mvc`
- make `ValidationHelpers` class `public`; already in `.Internal` namespace
- split `GetUserErrorMessageOrDefault()` in two; rename to `GetModelErrorMessageOrDefault()`
- fix some #YOLO wrapping
- aspnet/Razor#643 part 2: react to aspnet/Razor#653
- change test calls and delegates to include new parameter
- add tests specifically of the new feature
- add tag helpers using new feature to `TagHelpersTest` functional test
- note `HtmlEncoder`s used elsewhere e.g. in other `RazorPage` instances are unaffected
- i.e. changing one `RazorPage.HtmlEncoder` property only affects C# expressions in that page
Also simplify scope management, removing bizarre assertion when user does something odd.
nits:
- correct `// Act` and `// Act & Assert` content
- remove #YOLO wrapping
- #3386
- initialize comparison `HashSet` with unencoded values to ensure both are checked
- address perf and correctness issues in this code
- `context.Items[typeof(SelectTagHelper)]` entry read as `ICollection` but written as `IReadOnlyCollection`
- `IReadOnlyCollection` worse because it does not include `Contains()`, causing Linq use
- every `<option>` element recalculated the encoded values and created a `HashSet` to contain them
- add `CurrentValues` type to cache this `HashSet` in `context.Items`
- each `OptionTagHelper` created the additional `HashSet` even if `Value` was bound
This change adds a list of ApiRequestFormat objects to ApiDescription
object which include the content type and formatter for each supported
content type which can be understood by the action.
Computation is aware of the [Consumes] attribute via the
IApiRequestMetadataProvider metadata interface, and aware of Input
Formatters via the new IApiRequestFormatMetadataProvider interface.
This algorithm is essentially the same as what we do for
produces/output-formatters. We iterate the filters and ask them what
content types they think are supported. Then we cross check that list with
the formatters, and ask them which from that list are supported. If no
[Consumes] filters are used, the formatters will include everything they
support by default.
This feature and data is only available when an action has a [FromBody]
parameter, which will naturally exclude actions that handle GET or DELETE
and don't process the body.
- Validated that each of `TagHelperSample.Web`s sites work properly.
- Found that the `ConditionalCommentTagHelper` page wasn't working as intended due to never being updated after `TagHelperContent` encoding changes.
- Added content verification for the `TagHelper/ConditionalComment` endpoint only since it's static HTML.
#3530
Calling Flush[Async]() on the writer will NOT flush the stream.
Calling Flush[Async]() in Razor will flush both the writer and the stream.
Our normal flow will be to flush the writer, but not the stream. This
avoids chunking, but allows us to do a WriteAsync on the stream as part of
the call to FlushAsync. This is done to avoid a synchronous write due to
Dispose calling Flush on the writer, which needs to call Write on the
stream.
See issue for extensive background.
Adds the concept of an IAntiforgeryPolicy marker interface as well as
the ability to overide policy with a 'closer' filter.
Adds a new [IgnoreAntiforgeryToken] attribute for overriding a scoped
antiforgery policy.
Adds a new [AutoValidateAntiforgeryToken] attribute (good name tbd) for
applying an application-wide antiforgery token. The idea is that you can
configure this as a global filter if your site is acting as a pure
browser-based or 1st party SPA. This new attribute only validates the
token for unsafe HTTP methods, so you can apply it broadly.
- Check MVC services once at startup
- Make action selector sync
We've never really had a scenario for the action selector being async, it
just ended up that way. None of our extensibility here lets you do
anything async without replacing it wholesale, which we don't
recommend.
This change avoids a state machine allocation and a dictionary allocation
on the common case (no bound properties). Ugly? You bet. Worth it? Yeah,
seems worthwhile.
This is worth about 200 bytes/request - about 3% of allocated bytes in a
smallish API scenario.
- Always prepend with application name and let underlying `IStringLocalizerFactory` do the name gymnastics
- Use ExecutingFilePath instead of view name
- Trim off the file extension (so your resource doesn't have to have ".cshtml" in its name)
- Improved doc comments
- Added tests to cover ViewLocalizer behavior
- #3718
- #2767
- IHtmlLocalizer no longer derives from IStringLocalizer
- IHtmlLocalizer indexer now returns LocalizedHtmlString
- IHtmlLocalizer has GetString methods now that act the same as IStringLocalizer.GetString
- Made LocalizedHtmlString a struct to match LocalizedString
- Updated samples in response to aspnet/Localization#167
- Rename "ancestor" to "parent" for loc API
- Fixes some doc comments
- Fixed tests
- #3716
- #3612 subpart 2 of 2
- merge `ActionConstraintsWebSite`, `AntiforgeryTokenWebSite` and `ContentNegotiationWebSite` into `BasicWebSite`
- remove tests depending on `ErrorReporterMiddleware` in merged sites; not used in `BasicWebSite`
- delete `CompositeViewEngineWebSite`, `Microsoft.AspNet.Mvc.ViewEngines.CompositeViewEngineTest` covers cases
- delete `CorsMiddlewareWebsite`, soon to be in Cors repo (see aspnet/CORS#57)
- remove `SendFileMiddleware` from FilesWebSite, unused there
This change adds a base class for controllers to Mvc.Core that can be used
without a dependency on views, json, etc.
Most of the functionality on Controller moves to ControllerBase. I kept
the IActionFilter and IDisposable functionality on Controller since it's
not really a fit with the 'minimal' philosophy.
- Removes IExcludeTypeFilter
- Replaced with a property 'ValidateChildren' on ModelMetadata
- Teach ValidationVisitor to respect 'ValidateChildren' for enumerable
types.
The accessor for ElementMetadata can sometimes return null when
concurrently accessed.
This change solves this issue and simplified a bunch of lazy-computed
properties, by precomputing all type-related facets of ModelMetadata.
This also adds a ElementType property to ModelMetadata to maintain
the current separation of concerns as ModelMetadata is unaware of the
metadata provider.
- #3122 and #3123 (5 of 5)
- avoid concatenating values and wrapping them in an `HtmlString`
- switch from `string.Format()` to `AppendHtml()` in `LinkTagHelper`
- simplify `RazorPage` and `TagHelperContentExtensions` e.g. don't use a `TagHelperContentWrapperTextWriter`
- add some special cases in `UrlResolutionTagHelper`
- add `TagBuilder` and `StringHtmlContent` special cases to avoid `StringWriter` use when value is an `HtmlString`
- move `HtmlTextWriter` optimizations a bit lower and add missing checks for that type
nits:
- correct comments that incorrectly mention `HtmlString`s
- update comments in `JavaScriptStringArrayEncoder`
- part of #3123 (4 of 5)
- `LocalizedHtmlString` should not subclass `HtmlString`; now implements `IHtmlContent`
nits:
- clean up a few doc comments
- remove duplicate tests reported while testing these fixes in VS
This change removes the IActionContextAccessor as a dependency of
UrlHelper, and shifts UrlHelper to use a factory pattern. Consumers of
IUrlHelper should create an instance using the factory when needed.
This is the last part of MVC that has a dependency on IActionContext
accessor. As part of this change we no longer register it by default, and
treat it as an optional component.
This change removes the dependency of TempData on the IHttpContextAccessor
by creating an ITempDataDictionaryFactory abstraction. In general, no one
will replace the factory, it's just indirection.
This allows us to drop our dependency on IHttpContextAccessor, and move it
to the functional tests where we specifically depend on it.
The bulk of code churn here is to update tests that use TempData.
This change resolves#3512 and #3636 by removing 'magic' link generation
and adding an extension method to add routes to areas correctly using the new
pattern. This is pretty much exactly the same as how MapWebApiRoute works.
For site authors, we recommend adding area-specific routes in a way that
includes a default AND constraint for the area. Put your most specific
(for link generation) routes FIRST.
Ex:
routes.MapRoute(
"Admin/{controller}/{action}/{id?}",
defaults: new { area = "Admin" },
constraints: new { area = "Admin" });
The bulk of the changes here are to tests that unwittingly relied on the
old behavior.
- Prior to this change we were using `Linq` throughout our `TagHelper`s which resulted excess allocations.
- Also went through our data structures and changed some from `Enumerable<T>` to `IList<T>` where we were just not exposing the more accessible types (list).
- Changed several locations of `foreach` to `for` to remove the enumerator allocation.
#3599
This change simplifies IFormatFilter's API and removes the dependency on
IActionContext accessor.
The old API for IFormatFilter required computing state based on the
current request as part of the constructor, which in turn implied the use
of a context accessor. This isn't really needed. I didn't preserve caching of
the 'format' as that seems like an early optimization.
This is a companion change to make the RouteValueDictionary type more
promient. Using the concrete type here allows to avoid allocations in a
lot of common scenarios.
- #3123 (3 of 5 or so)
- react to rest of aspnet/Antiforgery@6a9b38d
- remove `HtmlEncoder` from localization requirements
- literal `hidden` is no longer HTML encoded (was a no-op anyhow)
- #3571 and part of #3123
- split `ContentViewComponentResult` in two
- add `HtmlEncoder` to `ViewComponentContext`
- remove use of `WebUtility.HtmlEncode()` and `HtmlDecode()`
nits: remove unused `using`s in files I had open
- #3428
- the `public` setter was not useful when this class is used as an attribute
- make property `virtual` to support other override patterns
- update existing test to use both override patterns
This change introduces ControllerContext for inside of Controllers, and
controller-specific extensibility points. ControllerContext carries with
it the model binding infrastructure needed to do all of the things that
controllers need to do.
- Removed `WriteTagHelperAsync` methods from `RazorPage`.
- Moved `WriteTagHelperAsync` tests into Razor since `TagHelperOutput` is now an `IHtmlContent`.
- Updated code generation test files.
aspnet/Razor#358
- `true` has the opposite meaning now but most changes are due to new parameters names in `IViewEngine`
- use name names in `Microsoft.AspNet.Mvc.ViewFound` and not found events
- remove `IRazorPage.IsPartial` and `RazorView.IsPartial`
- remove `IsPartial` properties from `Microsoft.AspNet.Mvc.Razor.BeginInstrumentationContext` and end events
- add parameter checks to `RazorView` constructor; instances are not retrieved from DI
nits:
- remove unused `cacheKey` parameter from `RazorViewEngine.CreateCacheResult()`
- correct duplicate test names in `RazorPageTest`
- also `...OnPageExecutionListenerContext` -> `...OnPageExecutionContext`
- `IRazorViewEngine.MakePathAbsolute()` -> `GetAbsolutePath()`
- set `IsPartial` on all `IRazorPage` instances
- improve consistency of methods in `HtmlHelperPartialExtensions`
- a couple unnecessarily passed `htmlHelper.ViewData`
- add missing tests of these extension methods
- restore parameter checks in `CompositeViewEngine`
- reduce `List<string>` and remove enumerator allocations in `CompositeViewEngine`
nits:
- correct a few comments
- use `<seealso/>`
- do not blindly use `FindPage()` / `FindView()` result in `Exception.Message` or returned results
- failure scenarios involve new `Any()` calls but rarely additional `List<string>()` allocations
- change `ViewEngine_ViewNotFound` resource to be consistent with similar errors
- remove trailing period at end of searched locations list
nit: remove remaining `null` checks of `SearchedLocations` in not found cases; never `null` then
- #3307
- relative paths are now supported in `View()` calls from components and view components,
`Html.PartialAsync()` and similar calls, and `RazorPage.Layout` settings.
- support absolute paths, relative paths, and view location lookups consistently / everywhere
- support view paths in `TemplateRenderer` e.g. passing an absolute path to `Html.EditorFor()`
- take a big swing at the `IRazorViewEngine` and `IViewEngine` interfaces
- split lookups (view names) from navigation (view paths)
- remove `Partial` separation; use parameters to set `IsPartial` properties
- correct `ViewContext` copy constructor and add unit test
- extend unit tests to cover relative paths
- fix existing tests to handle newly-required extension in an absolute path
- add functional test that chains relative paths
nits:
- remove some YOLO line wrapping
- `""` -> `string.Empty`
- Prior to this change the `<meta ...>` generated had a name attribute. As an alternative we're using the `content` attribute in conjunction with the `name` attribute to ensure compliance.
#3449
- wrong fix especially now that test encoders work as expected
- touch up `HtmlHelperTest` and `RazorViewTest` since test encoders are now consistent
- remove references to old `Microsoft.Extensions.WebEncoders.Testing` package
This reverts commit a9d5876cd9.
- but leave `FormTagHelperTest` and `ValidationMessageTagHelperTest` cleanup alone
nit: use `string.Empty` in `HtmlHelperLinkGenerationTest`
- #3140
- clone `MediaTypeHeaderValue` instance before updating it when content negotiation succeeds
- avoids changes to `MediaTypeConstants` properties and `OutputFormatter.SupportedMediaTypes` entries
- `MediaTypeHeaderValue.Clone()` does not exist in our DNX Core fork of this class
- in previous implementation, was called defensively rather than when required
- update `WebApiCompatShimBasicTest` functional tests to use `MvcTestFixture<TStartup>` everywhere
- #3140 blocked that final migration
- remove `TestHelper` since it's no longer referenced
nits:
- remove comments mentioning `TestHelper`
- correct spelling of "negotiation"
- test class can now use the `MvcTestFixture`
- #3139 part 3 of 3
- dump instrumentation data at end of `_Layout.cshtml`
- include `FilePath` in display
- compare against new `.html` resource
nits:
- normalize line endings to CRLF for consistency with other tests
- add `InstrumentionData` to avoid `Tuple`
- make `TestPageExecutionContext` class `private`
- remove newly-unused actions and associated `.cshtml` files
- remove tests already thoroughly covered in unit tests
- #3139 part 2 of 3
- make checks of `<form>` tag helper more consistent with others
- use a tag helper initializer but testing every variant; covered in unit tests
- remove `MvcStartupTest` and associated web site
- scenario covered elsewhere
- remove `PrecompilationTest.PrecompiledView_RendersCorrectly()`
- scenario covered elsewhere and shouldn't do I/O on checked-in files
- use `MvcTestFixture` instead of `TestHelper`
- part of #3139
- continue to test e2e interaction of `FlushAsync()` with sections, partials, et cetera
- remove `FlushPointTest.FlushBeforeCallingLayout()`
- `RazorPageTest.FlushAsync_ThrowsIfTheLayoutHasBeenSet()` unit test covers the scenario
- Added functional tests to validate data created from a `SelectTagHelper` does not impact following `<select>` tags.
- Also moved the new `SelectTagHelper` communication flow into `TagHelper.Init`.
#3347
- #3227
- much of change is to tests, creating and passing `ModelMetadata`
- updated `InputFormatterContext` to make `ModelMetadata` available to `JsonInputFormatter`
- walk `ModelMetadata` tree to get information about property with an issue
- add missing `null` checks in `ModelStateDictionaryExtensions`
- Changed all `GetChildContentAsync` calls to come from `TagHelperOutput` instead of `TagHelperContext`.
- Updated `ReaderAtEndOfFormTagHelper` to properly detect changes inside of a `Form`s body by calling the `Init` method.
- Add test to `RenderAtEndOfFormTagHelperTest`.
aspnet/Razor#571
The missing piece here is is that StringOutputFormatter needs to set the
ContentType so that it gets overridden. The problem is that the formatter
is likely called with something like application/json, but decides to
write a string anyway. So because we're saying 'yes we can write' we also
need to override what we're writing.
This change rewrites simple and safe Expression<Func<T, U>> expressions
into accesses to readonly fields. This allows us to cache the actual
expression and avoid repeatedly allocating and compiling it.
The rewrite is limited to cases where we know that the expression doesn't
capture, and where we support that kind of expression for evaluating
viewdata. In practice this means 'indentity' and property accessors are
allowed.
Ryan Nowak
Pranav K
Doug Bunting
N. Taylor Mullen
Brennan
Ajay Bhargav Baaskaran
Derek Gray
ryanbrandenburg
John Luo
jacalvar
Pavel Krymets
Kiran Challa
damianedwards
Hao Kung
David Fowler
Kristian Hellang
apxoht
herecydev
Cesar Blum Silveira
Kirthi Krishnamraju
Chris R
あまみや ゆうこ
Eilon Lipton
unknown