* Removes the need for a custom property when resolving static web assets from referenced projects
* Rolls in several improvements suggested by the MSBuild folks to improve performance.\n\nCommit migrated from 5335245b08
The code-generated TypeInference type resides in a custom namespace.
Any types that it refers to in user code must be qualified using the "global::" prefix
to avoid type \ namespace conflicts.
Fixes https://github.com/aspnet/AspNetCore/issues/12116\n\nCommit migrated from f0e09e4a97
- The core issue is that the Razor parser splits attribute values based on whitespace. Therefore, when it encounters `@onclick="() => Foo()"` it breaks it into three different tokens based on the spaces involved. This separation results in multiple adjacent classified spans for C# which is currently unsupported by WTE due to multiple seams overlapping. All that being said we have the opportunity to be smarter when generating attribute values that we feel can be simplified or collapsed; because of this in this PR I changed the `TagHelperBlockRewriter` phase to understand "simple" collapsible blocks and to then collapse them. In the future a goal would be to take a collapsing approach to all potential attributes and then to re-inspect each token individually at higher layers in order to decouple our TagHelper phases from what the parser initially parses.
- Added an integration and parser test to validate the new functionality. Most of the testing is from the fact that no other tests had to change because of this (it doesn't break anything).
- Added a new SyntaxNode method `GetTokens` that flattens a node into only its token representation.
aspnet/AspNetCoredotnet/aspnetcore-tooling#11826
\n\nCommit migrated from 80f1bc76a4
Don't even try to review this xD
This change moves RenderTreeBuilder to the .Rendering namespace and then
updates literally every component baseline.
\n\nCommit migrated from aa8624f46e
- Roslyn now default to 8.0 C# so it's no longer specified in a users project file. Because of this restriction we can no longer see the projects `LangVersion` when generating Razor files. Meaning, we can't conditionally generate C# 8.0 aware code because our Razor default is not C# 8.0. Therefore, when we detect that the C# lang version hasn't been provided we don't set a C# version and instead rely on what the default configuration for the Razor project is. In practice this looks like:
- MVC1.X = C# 7.3
- MVC2.X = C# 7.3
- MVC Latest = C# 8.0
- I thought about adding a feature flags variant for `RazorCodeGenerationOptions` but decided not to since C# features are all configurable options that are based on more than just the `RazorLangVersion`.
- Added integration tests to ensure that command line builds with implicit `LangVersion`s result in proper C# nullability handling.
aspnet/AspNetCoredotnet/aspnetcore-tooling#12594
\n\nCommit migrated from f039aa9354
- We would default to improper C# language versions when given non-numeric `LangVersion`s. For instance when given `CSharpLanguageVersion.Latest` we'd default to 7.3 which was inconsistent with how the rest of the C# world functioned.
- Added a test.
aspnet/AspNetCoredotnet/aspnetcore-tooling#11139
\n\nCommit migrated from a08d0b0673
- Updated instances of our code to no longer use reference type checks on `ISymbol`s. This was a new restriction added by Roslyn [here](https://github.com/dotnet/roslyn-analyzers/issues/2084). We also have a follow up AspNetCore issue [here](https://github.com/aspnet/AspNetCore/issues/12747) to bring these level of changes to AspNetCore.
- Had to pin `Microsoft.CodeAnalysis.Analyzers` in our VSIX projects in order to workaround version conflicts of `Microsoft.CodeAnalysis.Analyzers`. The version conflict is introduced from our dependency on `Microsoft.VisualStudio.ProjectSystem.Managed.VS`. It transitively depends on `Microsoft.CodeAnalysis.Analyzers` `2.6.3` where as our latest Roslyn bits transitively depend on `Microsoft.CodeAnlalysis.Analyzers` `2.9.4`. Tried updating to the latest, private, `Microsoft.VisualStudio.ProjectSystem.Managed.VS` but it also had the version conflict.
- Pinned our runtime bits to the latest public Roslyn NuGet package and pinned our tooling bits to the latest private NuGet package (both are compatible flavors of 3.3.0).
\n\nCommit migrated from 83d5e2c36f
- Prior to this a user could have private, protected or public parameters. We now require public and plan to update analyzers to enforce good usage of public component parameters.
- Removed private `ComponentTagHelperDescriptorProvider` workaround. meta data workaround since we no longer operate on private parameters.
- Updated existing tests to not have private parameter setters. I missed these in my last pass.
- Added two new tests to validate parameters which are private or have private setters are ignored.
aspnet/AspNetCoredotnet/aspnetcore-tooling#12291
\n\nCommit migrated from bb543aa166
- Directives utilize helpers to add to colorization errors, etc. in Razor files. These helpers typically look like `MyType __something = default(MyType)`. The issue with this approach is that `default(MyType)` can be `null` yet `MyType` is a non-nullable type. To workaround this restriction I added nulllability forgiveness operators to the type and member directive tokens (the only two that could possibly generate `null` based output).
- Updated baselines to reflect new forgiveness operator.
aspnet/AspNetCoredotnet/aspnetcore-tooling#10919
aspnet/AspNetCoredotnet/aspnetcore-tooling#11478
\n\nCommit migrated from d0617ed578
- Updated `BuildVariables` to lift the `Microsoft.Net.Compilers.Toolset` package version into C# so we can pass it as a parameter to our testapps when building them.
- Flowed the `MicrosoftNetCompilersToolsetPackageVersion` to transitive testapp projects on restore. This enables `build.cmd` to properly pre-build our `testapp` projects with the latest version of Roslyn.
- Added several conditions to detect if a user is attempting to independently build a `testapp` at which point we fallback to using the SDKs Roslyn compiler toolset version.
aspnet/AspNetCoredotnet/aspnetcore-tooling#12268
\n\nCommit migrated from 57a72ecc46
* Correctly set UpToDateReloadFileTypes
UpToDateReloadFileTypes was incorrectly set in the preview6 DesignTime targets and this was worked
around in the SDK's targets. Unfortunately the updated copy of DesignTime targets (current as of 7/13)
also has a bug.
This fixes the issue and adds some tests to verify we're correct with the past, present & future.
Fixes https://github.com/aspnet/AspNetCore/issues/11873
\n\nCommit migrated from 525d0c1ad3
I noticed this while debugging something else in VS. It looks like the
service hub is using camelCase, which wasn't covered with a diagnostic
in our tests. As a result we would fail to read the data from the OOP
host, and fall back to in-process tag helper discovery.
\n\nCommit migrated from 63b25dbd2d
* Fix weakly-typed component bind
The globalization change regressed this. For weakly typed component bind
we don't want to do any conversions, we need the values to match
exactly.
Note that there's no compiler-generated type checks because its
weakly-typed.
* pr feedback
\n\nCommit migrated from 2d74721815
- This is a temporary workaround until we're able to get on the latest SDK which includes the latest Roslyn bits. Without them we can't properly publish testapps in a standalone fashion without investing in temporary infrastructure.
\n\nCommit migrated from 21fcd9c2f5
- Verified the fix with VS16.2-preview2 + latest CLI with modified `Nullable` entry on disk
- Updated tests.
- Note: This change will not pass the build until we have a new CLI that has an updated Roslyn that understands `Nullable`.
aspnet/AspNetCoredotnet/aspnetcore-tooling#10218
\n\nCommit migrated from fee9c35bee
* Update ref assembly APIs
There are some impactful changes here for the compiler that are part of the API review.
The next few commits will implement all of the required changes.
* React to API review + globalization changes
* Updates to generated baselines
* Add missed file
* Update src/Razor/src/Microsoft.CodeAnalysis.Razor/EventHandlerTagHelperDescriptorProvider.cs
Co-Authored-By: Steve Sanderson <SteveSandersonMS@users.noreply.github.com>
* more baseline updates
\n\nCommit migrated from d955be5013
* Make some tweaks to the default content globs
- Don't publish/copy launchSettings.json
- Make sure static content and configuration work with single file turned on
- Added test
\n\nCommit migrated from 63e8681fff
* Modify tag helper infrastructure to configure case sensitivity
* Cleanup and added more tests
* More cleanup and added necessary diagnostics
* More feedback
* Added more tests
\n\nCommit migrated from 20b8286751
- Updated existing tests to use `public` accessibility for parameters. As part of this also updated baselines.
- We now recommend that `[Parameter]`'s should be `public`. In the future this will be a requirement.
aspnet/AspNetCoredotnet/aspnetcore-tooling#8825
\n\nCommit migrated from 2ceb597f82
- Without this any diagnostics from Analyzers will be doubly reported. One for the declaration compilation and one for the view compilation.
aspnet/AspNetCoredotnet/aspnetcore-tooling#8825
\n\nCommit migrated from 6d115d460b
- Added a feature flag to ensure this functionality does not impact pre-3.0 applications. This change is in fact a breaking change because some less-common scenarios will no longer be valid. For instance, `@Name!.Hello`used to treat the `!.Hello` as markup, now it will be treated as part of the expression.ExpressionChunkGenerator
- Updated parser flags tests
- Updated parser tests
- Updated the C# 8.0 integration tests to include the new nullable forgiveness scenarios.
aspnet/AspNetCoredotnet/aspnetcore-tooling#11188
\n\nCommit migrated from 625c605607
This fixes a regression in the new `@ref` support, where the build will
fail during the declaration pass.
The problem is that the EliminateMethodBodyPass runs really early and
wipes out the usage of `@ref` before we have a chance to see it. The
solution is to make it run really late. This makes sense because we
basically just use this pass to do cleanup.
\n\nCommit migrated from 09b50151ef
Adds the code generation support for culture + bind.
Some additonal changes are needed for the compiler support for bind, we
need to pass the culture into the code that unwraps the value as well.
There's no need to eagerly push these changes into the runtime (can wait
for the compiler change to go in) because nothing in our runtime tests
use the new feature.
\n\nCommit migrated from 6aad9748e8
Adds the ability to specify that an `<input type="..." />` for some
specific value of `type` maps to the invariant culture and/or provides
a default format.
The primary examples of usage would be the number field which wants to
use invariant culture, or datetime which wants to use invariant culture
and a format string.
\n\nCommit migrated from 60f2d563a6
Updates our Components shim to use the ref assembly code. This is just
cleanup and will help us keep up-to-date with API review changes since
we can just copy-past the files.
I made two modifications to make this simple.
- Removed our built-in components
- Removed the HTTP stuff
This makes it easier for us to maintain the shim. You can see cases
where I had to update tests because they diverged from what the real
APIs look like (no product bugs).
I'd still like to improve this workflow so suggestions wanted xD
\n\nCommit migrated from 8517be2bac
This change makes @ref define a field for you by default. We didn't do
this originally because we weren't sure if the compiler generating a
field would always be what you want, and more importantly,
we didn't have the expressiveness to add an opt-out.
Now that we have directive attributes, it's easy to create an opt-out
for the field generation. The special thing about it is that where
you opt-out has to be static. For this reason the @bind:suppressField
attribute cannot be used with a value, you either have the flag or not.
Additionally, we don't support automatic field generation for
generics, because usually we can't write the field name. If we
want to in the future could make this work when the generic
type arguments are specified. So in the case of generics,
you have to opt-out, and you will get a diagnostic
that tells you so.\n\nCommit migrated from dadf9faf22
Currently FindAssembliesWithReferencesTo produces a stacktrace when P2P builds fail - for instance
if the referenced project has a compiler error. This changes it to use LogErrorFromException.
In addition, there was a typo in the property that disabled the feature.
Fixes https://github.com/aspnet/AspNetCore/issues/11226\n\nCommit migrated from 87c12a116c
- add feeds that SDK no longer provides automatically when building test apps
- add another allowed build warning in some integration tests\n\nCommit migrated from cfd4698381
* Add TagHelperDescriptorProvider
* Remove bad life-choice
* Add some AddRange love
* Add splatting lowering and codegen
* PR feedback
\n\nCommit migrated from 9c8dd1cae7
* Allow @bind:event override when used with @bind
* Added more tests to cover standalone @bind-value=... case
* fix
\n\nCommit migrated from a399bd072e
- It looks that when we first did VCTH we never put in the effort to reduce our footprint on available parameters for `ViewComponentTagHelper`s. I went ahead and added several underscores to the generated fields/parameters to reduce collisions.
- Updated tests to reflect the new codegen.
aspnet/AspNetCoredotnet/aspnetcore-tooling#8645
\n\nCommit migrated from b545a4de75
- WTE needs a way to detect if a `TagHelperDescriptor` is a component based `TagHelper` in order to turn features like validation, completion etc. on/off.
- Updated tests to add verification points for the new `IsRazorComponentTagHelper` method.
\n\nCommit migrated from 6fcd12e60b
* Adds support for publishing static web assets
* At publish time, it copies all the referenced assets from referenced
projects and packages into their final locations.
* Automatically pack static web assets for consumption
* Generate Microsoft.AspNetCore.StaticWebAssets.props and pack it into
build\Microsoft.AspNetCore.StaticWebAssets.props
* Generate `<<PackageId>>.props` and pack it into `build\<<PackageId>>.props`
importing Microsoft.AspNetCore.StaticWebAssets.props
* Generate `<<PackageId>>.props` and pack it into buildMultiTargeting\`<<PackageId>>.props`
importing `build\<<PackageId>>.props`
* Generate `<<PackageId>>.props` and pack it into buildTransitive\`<<PackageId>>.props`
importing buildMultiTargeting\`<<PackageId>>.props`
* Pack all the static web assets from the current project into `staticwebassets\**`\n\nCommit migrated from 87817bed3e
* Define static web assets in the current project
* These are assets under the wwwroot folder.
* By convention, the base path for these assets is `_content/<<PackageId>>`
with spaces and dots removed and all characters lower-cased.
* Retrieve static web assets from referenced projects\n\nCommit migrated from 50646aae64
* Specify UpToDateReloadFileTypes
Specify a list of file extensions that VS will monitor for file changes to reload the app.
Includes .razor and .cshtml files.
\n\nCommit migrated from 9a964127bf
This came up in the context of discussion around aspnet/AspNetCoredotnet/aspnetcore-tooling#5071.
Browsers use a 'first attribute wins' rule for evaluating attributes and
building the DOM - though duplicate attributes are considered to be a
spec violation.
Blazor wants to use 'last attribute wins' rules because that's most
sensible when you consider evaluation order and splatting.
This means that we have to have the markup block pass do the same thing
as the Blazor runtime when duplicates appear. However, since this is a
spec violation I'm proposing that we just make it an error (like we do
for malformed tags). There's no useful purpose for duplicate attributes
since browsers ignore it.
Since we're not going to allow this for elements. Let's go all of the
way and block it for components as well. And while we're here, lets
block some of the invalid cases where users try to mix onchange and
bind. These don't work today, so we should make them an error. If we
decide to add support for this case to the runtime that would require
compiler work anyway.
\n\nCommit migrated from 0c59ca5f74
* Directive attributes part 1: Support parameters in bound attributes
* update
* Fix test
* feedback
* Updated design
* Do case sensitive comparison
* more
* Bug fix
\n\nCommit migrated from 9bbf240948
* Imports static assets from packages containing custom msbuild
targets defining a StaticWebAsset item group.
* Generates an embeds a manifest into the application assembly
that contains a list of paths to the content roots of the
assets defined in the packages custom msbuild targets.\n\nCommit migrated from 8e70013b70
* Move the globs from the project system into the Razor SDK on a separate file that can be imported by itself.
* Import the file conditionally based on the EnableRazorContent property\n\nCommit migrated from 62e33dac1d
The intention here would have been to add to the existing GetCopyToOutputDirectoryItemsDependsOn property, defined in Microsoft.Common.Targets. Instead what was happening is that it was being completely overwitten with the _RazorGetCopyToOutputDirectoryItems entry, given that $(GetCopyToOutputDirectoryItems) doesn't exist as a property and must have been a typo (it exists as a task, not a property).
An effect of this bug was that the `AssignTargetPaths` target is not always executed before the `GetCopyToOutputDirectoryItems` target, which should explicitly depend on it. `AssignTargetPaths` is responsible for populating the `ContentWithTargetPath` itemgroup. Left empty, it appears to `GetCopyToOutputDirectoryItems` task that there are no Content files to copy to the final output directory, when in fact there are.\n\nCommit migrated from 4597bc3d55
- This is a pre-requisite to properly warnings users about TagHelpers in code blocks. With this change VS indicates that there's an asynchronous operation happening in a code block but because the code is auto-generated it happens behind the scenes and therefore always gets logged as a line 1 error (unusable). Therefore, to fix this issue there will be a partner PR that will go out in AspNetCore that adds analyzer errors for cases where `__tagHelperRunner.RunAsync` exists in code blocks that do not return `Task` and are not marked with `async` properly.
- Updated test files to reflect the new TagHelper codegen at design time.
- Note: This doesn't solve the `~/` case that will be in a follow up PR.
aspnet/AspNetCoredotnet/aspnetcore-tooling#8630
\n\nCommit migrated from 17af3efccd
- This adds support for C# single line variations of: `if`, `for`, `foreach`, `do`, `while`, `lock` and `using`.
- Turns out the existing parser had 99% of the support for these scenarios already. Therefore, in this change set I went ahead and added exhaustive tests to verify things worked end-to-end.
- Added a restriction to single line markup to not allow markup in single line control flow statements. Updated resx to provide a better error message for cases when users do use markup in single line control flow statements.
dotnet/aspnetcore-tooling#9637
\n\nCommit migrated from cf474c97aa
- Diagnostics were not being raised to the `RazorSyntaxTree` and weren't being ordered on final output. This resulted in some of our tests missing the fact that certain cases were generating errors.
- Made all three phases of Razor parsing order their diagnostics.
- Added `GetAllDiagnostics` methods to `SyntaxNode`s to be more consistent with IR documents.
- Updated test files. In some cases new errors were found because we're now lifting them to the `SyntaxTree`, in most others the errors are re-ordered.
**Note: In end-to-end scenarios diagnostics were not hidden, only unordered. The IR phase would find nested/hidden documents and lift them to the IR/C# documents.**
\n\nCommit migrated from ef31a9683b
Javier Calvarro Nelson
Pranav K
Ajay Bhargav Baaskaran
N. Taylor Mullen
Ryan Nowak
Ryan Nowak
David Fowler
Doug Bunting
Steve Sanderson
Nate McMaster
Ajay Bhargav Baaskaran
Alex Lorimer