Improve the grammar on a diagnostic.
Add further tests requested by reviewer.

as well as other tests requested in code review.

Plus a couple of other minor changes per code review.

Make it clear that a cast is valid by tightening the type of the value used to initialize its storage.

Adjust scope of locals declared within a Yield Return and Local Declaration Statements according to the latest design.

Also eliminates some cases of cascaded definite assignment errors.
Fixes #13219
Fixes #13009
Fixes #12996
Fixes #10446

Adjust scope of locals declared within a Yield Return and Local Declaration Statements according to the latest design.

Related to #12597.

made checksum tree builder to only use green node not only to cache b…

I will add those back if necessary when I do more aggressive perf optimization

Don't allocate a lambda each time we get a hashcode for a named type.

fixed race in analyzer driver

push file read to Background thread.

fixing up stack trace logic as per Cyrus' suggestions

existing diagnostic tagger didn't check workspace diagnostics belong …

add more cancellation checking in LB code path.

Fix some failing test on non-english machines

Add support for embedding VB source in portable pdb

Code Model: Implement CodeNamespace.Comment and CodeNamespace.DocComment properties

* test perf.groovy file

* Pull down CPC

* Changes machine pool name

* Set env-variable for willow install

* Generate PR job too

* address feedback

* Remove diagnostic

* Correct regex

this should fix /workaround one particular case of dead lock but it won't fix the root cause of this where DispatcherSynchronizationContext Wait allowing re-enterance of COM call which, if re-entered code ever use Wait/Result, causes VS to dead lock.

alternative to this fix is using Dispatcher.CurrentDispatcer.DisableProcessing(), but since workspace event is so busy code path, didn't want to play with global state (resource) all other people are using.

code path where code fix uses didn't check cancellation itself but just pass it down to workspace/compiler. but that will make it to not cancel if all diagnostics are already calculated and cached. so it will always run to end and then right after that, code fix service will throw cancellation exception.

by adding some cancellation here, we can prevent us from doing unnecessary work.

http://source.roslyn.io/#Microsoft.CodeAnalysis/DiagnosticAnalyzer/AnalyzerDriver.cs,1699

var declarationAnalysisData = compilationData.GetOrComputeDeclarationAnalysisData(
decl,
allocateData => ComputeDeclarationAnalysisData(symbol, decl, semanticModel, shouldExecuteSyntaxNodeActions, analysisScope, allocateData, cancellationToken),
cacheAnalysisData);

GetOrComputeDeclarationAnalysisData caches declarationAnalysisData only based on declaration, but the data it caches (DeclarationAnalysisData) depends on more context than declaration node. such as analysisScope. more specifically FilterSpanOpt or executeSyntaxNodeActions flag.

due to that, it now has a race where depends on what analyzer are loaded to VS first, some syntax node action works, or some other one doesn't.

the fix I made here is removing one of those context when caching the result

that caused host workspace's diagnostics to be removed when preview workspace's diagnostics are removed since two shared same documentId and projectId.

the code already had the logic to check what workspace the diagnostics belong to, it just happen to check that after removing diagnostics. fix is moving the checking before removing diagnostics.

made checksum tree builder to only use green node not only to cache but to build tree. this should prevent checksum builder to realize all red nodes for solution just to create checksum tree.

this change also include using specialized small dictionary to prevent creating more expensive dictionary.

another optimization can be done is moving checksum data into solution itself so we don't need to have side collections to map between solution green node to checksum node. but I will do that once I get to point where I do more aggressive perf optimizatino to reduce memory foot print if that is necessary

This change properly implements the CodeNamespace.Comment and CodeNamespace.DocComment properties. Historically, C# would throw an exception if either of these properties were accessed on a CodeNamespace, but VB allowed them to work. With this change, the properties will work for *both* C# and VB (even though DocComment doesn't make much sense). Technically, this is a breaking change for C# in the sense that an exception will no longer be thrown. However, I expect that it will generally make life a bit easier for Code Model consumers by keeping the implementations consistent between C# and VB.

* Pull down CPC

* clear files before testing

Remove file that exists on disk, but isn't referenced by any Roslyn project.

set references to some assemblies to `<Private>false</Private>`

Fixes #13062
Closes #13091