This change implements version tracking the inputs and outputs of
generated code.
Version tracking is still best-effort - meaning that in some cases a
perfect system could avoid doing more work. However, since we base the
versions off of all of the inputs, we now that the guarantee that code
generation operations that happen 'out of order' will always result in
the newer inputs generating the newer outputs.
Fixes: https://github.com/aspnet/Razor/issues/2650
This change adds mock ups of the interfaces that we've been designing as
part of Razor FAR as well as the implementations. This isn't wired up to
anything yet in this PR, but the basic functionality here is stable
enough for us to stabilize and review.
For now we have the interface definitions in the Razor codebase until a
build of Roslyn is available with these definitions + IVT for us to use
them.
This builds support for tracking the effect of changes to imports on
other documents, and completes our model for being able to keep
generated code up to date.
Addresses a blocking issue for FAR of types when used in user-code in a
directive. The FAR infrastructure is skipping over the directive code
because it's mapped to `#hidden`. As you can see in the code, the token
provided by the user is already included in the projection mappings.
I think we didn't do this before because we didn't expect this code to
need line numbers - it's not really debuggable, and design-time codegen
doesn't happen when you build the project.
I think it's OK for now that we don't line-map (or include) directives
based on view imports. If you trigger FAR on an `@inject ...` in an
import for instance, you'll find the reference for the view import file.
That seems pretty good, and the only cases I can really imagine it being
broken would be for go-to-definition (within a Razor view). Lets revisit
in the future based on feedback.
- Prior to this project changes would trigger re-parses which would then be thrown away because source versions were identical.
- Added test to verify new SetOutput behavior.
aspnet/Razor.VSCode#184
The Blazor test infrastructure made a copy of the code in Razor and then
added some more features. This change backports the features needed for
the style of test we're using in Blazor.
I updated the MVC integration tests to use the new style, but I think
there's limited value in trying to rev all of the old tests.
One feature in particular that I removed from the old infrastructure was
the automatic inference of imports based on the file system. This
feature was wierd and doesn't parallel how these features work in
actuality. It's easy and more natural to test imports in new style
tests.
Loading multiple versions of a task assembly result results in MSBuild on .NET Core to fail.
Addressing this by moving the tasks in to the Sdk and renaming it. This should avoid
immediate issues for a 2.1 and 2.2 project co-existing and future proofs 2.2 and later.
Fixes https://github.com/aspnet/Razor/issues/2553
- Prior to this we had a `BackgroundDocumentGenerator` that would constantly be updating generated code containers. With this changes we've changed the details in how a `GeneratedCodeContainer` can be mutated. It can now be touched from any thread and is updated when an underlying `DocumentSnapshot` has available content. However, if a generated output comes through that's older then the last seen output we no-op.
- Removed `BackgroundDocumentGenerator` SetOutput logic.
- Updated tests to react to new behavior of `GetGeneratedOutputAsync`.
- Added new test to verify getting generated output results in the setting of the host documents output.
- The issue was that the `executeChildContentAsync` call was stopping the invocation of the `_endTagHelperWritingScope`. Therefore, we'd never finish the TagHelper content scope and all following content would be ignored.
- Added two tests to validate the new functionality.
#2561
* Auto generate CSharp syntax tree nodes
- Auto generated syntax nodes, visitors and factories
- Renamed RazorComment to RazorCommentLiteral to avoid a conflict
- Modified the Visit() contract
- Added new SyntaxKinds
- Regenerated baselines
* Updated the tokenizer and parser to operate on SyntaxTokens directly
- The tokenizer and parser now operates directly on SyntaxTokens (Green)
- The SyntaxToken(Red) is now created when the Span is built with the correct parent and position
- All other passes that run after the parsing is complete(TagHelperRewriter etc) will operate on the Red tokens.
- There is now only one type for all SyntaxTokens. They are differentiated by their SyntaxKind.
- Added equivalence checking for tokens
- Updated test code to react
- Regenerated baselines
Initial infrastrucure for a better html parser
- Added new syntax node types that follow Roslyn's red/green pattern
- Modified the html text node in the current syntax tree to use the new
nodes
- Updated test infrastructure
- Generated baselines
- Migrated the completion item source provider and the legacy directive completion provider to use the new service.
- Cleaned up duplicate tests that were both verifying common completion functionality.
- Ensured that the legacy `RazorDirectiveCompletionProvider` did not result in additional Razor assembly loads when in C# scenarios.
#2530
- Kept the same behavior as we previously had with Razor directive completions.
- Attempted adding additional functionalities such as lighting up Razor directive completion when completion was invoked on top of Razor directives (non-C#) but ran into issues involving the core HTML editor not consuming the new completion APIs yet. That's something we'll have to re-visit once they move to the new completion APIs.
- Added tests to validate all aspects of new completion APIs.
- Made completion provider turn on and off based off of feature flag.
#1743#1813
- Some time spans were too aggressive for parsing a Razor file on a slow environment.
- Fixed an obvious race where we'd expect 0 pending notifications when it was possible to be in the process of clearing notifications.
#2506
For some inexplicable reason timestamps returned as part of the test often show up
as being different resulting in test flakiness. We've manually verified that this
is not a product issue and builds are correctly incremental on xplat. See #2219
for past discussions.
Fixes https://github.com/aspnet/Razor/issues/2503
- When a surrogate pair is about to be split down the middle we reduce the size of our split by 1 character. This way we split right before a surrogate pair. In the case of zero width joiners, if we split on a zero width joiner we still render valid bytes because the zero width joiner by itself can stand alone.
- Added tests for the various split cases.
#2470
Ryan Nowak
N. Taylor Mullen
Nate McMaster
Ajay Bhargav Baaskaran
Pranav K
Ajay Bhargav Baaskaran
Artak
Doug Bunting