This is a perf improvement of about 500ms for our razor code generation
benchmark on my dev box. That's about .8% of the overall execution time of
this benchmark.
This change will remove a bunch of unnessary allocations from the
parsing/code-generation path, and should improve responsiveness.
For reference Enum.HasFlags performs boxing of the enum value, and then
does a type comparison to see if the types are the same. This is
significantly more costly than a normal bitwise and comparison, and it
results in allocations.
These types are allocated for every state transition in the parser (State)
or every character that's read (character reference) and are fairly
compact. Turning them into structs will remove a significant number of unnecessary
allocations in the parser.
For our razor code generation benchmark, making these changes yields about
500-750ms of speedup (out of 40000ms) or about 1%.
- This registration system accepts tag descriptors that it then uses to feed requests for tag helpers.
- Added some infrastructure pieces that are used to build up valid tag helper descriptors.
#70
GetSourceLocation is frequently called to determine the location mappings
between the original document and the generated code.
The old implementation did a number of ToString and replace operations to
simplify the math on tracking the position - which put it front and center
in our performance measurements - about 25% of all execution time in a
sampling profile of our perf test.
The new code tracks position as code is written, and avoids allocations.
After these changes GetSourceLocation doesn't show up in the profile.
This extensibility point needs to be an override and not a pure visitor due to how the class declaration is created (only 1 right way to write the class).
#76
Removed all System.CodeDOM related code from net45. This involved modifying some existing code paths and most of all reducing the complexity of all of the CodeGenerators. Added some factory creation methods to build CodeBuilders. Ran code analysis and made adjustments to pass analysis.
Added newlines inbetween ending line pragma's and code. Without the extra line the document does not format correctly. Separated expression and statement padding functionality. Statements need to have 1 less padding to account for the transition. Changed how runtime and design time code generates to enable accurate debugging experiences in runtime and functional formatting experiences during design time.
This allows for users to write "@using System;" and still have proper intellisense and mapping. Also removed some legacy code that I came across when running tests.
.NET FX Team will not be adding Char.GetUnicodeCategory because it breaks layering. We cannot have a dependency form System.Runtime.dll to System.Globalization.dll.
Added a CSharpPaddingBuilder based on the existing
CodeGeneratorPaddingHelper to allow accurate padding within the generated
C# files. Also created tests based on the existing PaddingTest tests to
verify padding functionality.
This involved adding more boiler plate to the base of the visitor classes such as a Writer and a Context. By following this contract for new visitors we'll ensure that visitors have enough information to add new features to Razor.
Without the initial code tree acceptance you end up stack overflowing. Also removed a validation check that ensures that there's always a base type (not true).
This addition will eventually replace CodeDOM to allow for the k10 project to build. It is a new type of codegeneration method that is highly extensible.
Pranav K
Ryan Nowak
N. Taylor Mullen
David Fowler
Brice Lambson
Yishai Galatzer
Andrew Peters
Nick Guerrera