Change the display name of the roslyn analyzers to "Roslyn Analyzers" to avoid conflicting with the "Roslyn Diagnostics" window.

Bonus: move the project into the folder hierarchy.

CR: srivatsn; kevinpi (changeset 1388791)

Allow PE and PDB streams to be arbitrary writable streams. If the specified stream doesn't fulfill implementation specific requirements of the PE/PDB writer use an intermediate in-memory stream and write its content to the user specified one at the end of emit.

Fixes #349.
 (changeset 1388751)

Add a better description for people who encounter the diagnostic about prefixed verbatim crefs. (changeset 1388376)

We also add all of the relevant optimizations requested for current and future
formatting performance improvements. We now use full overload resolution
(using a new binder type specialized for synthesizing invocations) to
select an appropriate string.Format method or format factory method, which
enables it to handle params, generics, arbitrary argument conversions, etc, and we
optimize the special case of no fill-ins by processing the {{ and }} escapes,
producing a literal in the generated code. (changeset 1388224)

Within the compiler layer, we want to access properties like Category, DefaultSeverity, IsEnabledByDefault, etc., off of Diagnostic directly rather than going through Diagnostic.Descriptor. Roslyn diagnostic RS0013 enforces this.

However, it fires within Diagnostic itself, even though this is the one place where accessing Descriptor is legitimate. This change adds suppressions to these occurrences, but the diagnostic should be updated to not fire in Diagnostic itself. A bug has been filed to track that work, and updating the Roslyn solutions to use new toolset & diagnostics packages with the fix.
 (changeset 1388110)

Index expressions in the index initializers should be evaluated exactly once.
In a case of a nested initializer, we were evaluating as many times as we have members in the RHS. That resulted in a number of bugs reported.

Working on this fix revealed some more issues with params and named indices not being supported properly.

Fixes #449
 (changeset 1388102)

Bug 1022828: Buffer overrun in PdbWriter - off-by-one error, we were writing to the SymWriter's stack beyond a reserved buffer.
EnC apply tests can run 32bit.
Resource tests were incorrectly converting a pointer to 32bit integer. (changeset 1387932)

Add a test for 981208. The new cache gets collected when you call Workspace.Dispose() and remove references to the workspace and things that hold onto it, so we just need regression test. (changeset 1387925)

Call WithDocumentText when fed with the same text sometimes doesn't return the same instance of the Document because TextSource, a WeakReference, sometimes loses the reference and the perf optimization to return the same instance of the Document is not possbile. (changeset 1387876)

We shouldn't crash just because a PDB is missing information for a specific method.

It looks like the other callers of GetMethodByVersion handle null appropriately.

CR: ChuckS; KevinH (changeset 1387853)

***NO_CI***
 (changeset 1387627)

We weren't handling a number of the different types of state machine fields, and the Kind returned for hoisted, synthesized local fields ("$S" prefix) looked wrong (returned "HoistedLocalField" instead of "HoistedSynthesizedLocalField") (changeset 1387592)

***NO_CI***
 (changeset 1387571)

This changes EquatableAnalyzer to use a OfType call instead of Cast.  Using Cast here is incorrect and is causing an exception to be thrown in our toolset.

     Bug 1097052
***NO_CI***
 (changeset 1387470)

     The most immediate issue here is cancellation of the CancellationToken passed to AsyncQueue<T> will unconditionally call TaskCompletionSource.SetResult for the main Task value in the queue.  This cancellation can happen long after or in parallel with the completion of AsyncQueue<T>.  When this happens an exception is thrown.  This exception is causing the VBCSCompiler process to crash because it expects cancelling a token to be a safe operation.

     Longer term there are other issues in AsyncQueue<T> that need to be fixed (several race conditions).  Once this is checked in I willmove 1097123 to a Pri 1 and track the remaining fixes with that.   (changeset 1387407)

Changed SymbolEditor to use SyntaxEditor for editting declarations. (changeset 1387404)

NameSyntaxComparer was causing stack overflow due to infinite recursion. (changeset 1387310)

This is consistent with the spec and Dev12.
***NO_CI***
 (changeset 1387266)

***NO_CI***
 (changeset 1387232)

* Add 'when' keyword recommender to the completion list
* Ensure that the completion appears inside exception filters.
* Add 'when' keyword span to keyword highlighting for try..catch statements.
* Ensure that parenthesis simplication works catch filter expressions
* Add tests for 'when' classification (it already was classified correctly) (changeset 1387214)

This adds a per-rule Title for all C# and VB compiler warnings that have message strings.  These Titles will show up in the ruleset editor.

This also includes Descriptions that show up upon expansion for a set of high-hit C# warnings.  I added Descriptions for high-hit warnings per MSDN.  The work is still pending to add the equivalent VB warning Descriptions.

The principle was to basically restate the base Message string without its fill-ins as the Title, swapping in general words such as "method" and "type".  Often, just removing the fill-ins and quotes was enough, as the surrounding context already said what the code element was.  Sometimes the message needed to be reworded a bit.  I also left out any further clauses (such as "Consider ..."), except for shortened versions necessary to disambiguate similar warnings.  For now, I'm leaving such clauses in the messages themselves, using Description for longer-form explanations - we can see if we get feedback there.

This should be the content work needed to ship the ruleset editor without blank titles.  There was one key bug remaining (now filed) where rules kept around solely to make previous /nowarns and #pragmas happy still show up in the ruleset editor.

Remaining work:
* Descriptions for VB compiler warnings
* Titles/Descriptions for C#/VB compiler errors
* Categories for VB/C# compiler warnings/errors (changeset 1387213)

Stop throwing Unreachable exception in CalculateLocalSyntaxOffsetInSynthesizedConstructor, return -1 instead.
***NO_CI***
 (changeset 1387208)

***NO_CI***
 (changeset 1387165)

Carefully restructure AsyncRewriter.CreateMethodBinder so that it works for both the compiler and the EE.

CR: chucks; ngafter; alekseyt (changeset 1387147)

* Fix up existing IDE support for C# string interpolation to account for the new syntax. The new model makes the implementation much simpler!
* Fix formatting of alignment and format specifiers
* Unskipped all IDE tests that were previously skipped due to the C# string interpolation changes (changeset 1387125)

Before navigating to a symbol defined in a given document, check whether a related linked or shared document is the active context document and use that document instead. For regular files (non-shared and non-linked) and closed linked files, this is always the provided document. For open linked files and open shared files, the active context is already tracked by the workspace and can be looked up directly. For closed shared files, the document in the shared project's SharedItemContextHierarchy is preferred. (changeset 1387121)

Make sure we use the same CodeAnalysisRuleSet across all configurations of a project. (changeset 1387120)

Disabling a few suites which are exhibiting flaky behavior in the lab.  DevDiv 1095079 tracks re-enabling them. (changeset 1386998)

***NO_CI***
 (changeset 1386846)

- Group code fixes offered in the light bulb by the issue being fixed (as opposed to by the provider that supplied the fix). The order of the offered fixes should mostly remain the same as before - however, they will now be grouped differently - in most cases this will result in fewer groups in the overall list + this should make it easier to understand where fixes for one issue end and next one begins (in cases where we have fixes for more than one diagnostic on a given line)... (This addresses bug 1077181)

- Stop prioritizing fixes based on whether they are fixing a warning / error / hidden diagnostic. Fixes are already being ordered (based on ExtensionOrdering as well as based on proximity to cursor) - we shouldn't allow the priority setting to interfere with this ordering. We now use priority only for ordering fixes above refactorings and for ordering everything above the suppression fixes. This should also fix the issue where the rename tracking code fix would always show up at the bottom of the list below GFU fixes. (This addresses bugs 1090799 and 1074182) (changeset 1386745)

Ensure that GetSemanticInfo() returns the correct members for the LHS of a conditional access expression.

The syntax node for the LHS of a conditional access expression appears in two places:
- Inside its own BoundExpression, tracked by BoundConditionalAccess.Receiver
- Inside the BoundConditionalReceiver that is attached to the RHS of the conditional access expression, tracked by BoundConditionalAccess.AccessExpression

Or, diagramatically (let "SyntaxNode" be the syntax node for the LHS of the conditional access expression):

      BoundConditionalAccessExpression
      /                              \
BoundExpression                    .....
      |                              |
  SyntaxNode               BoundConditionalReceiver
                                     |
                                 SyntaxNode

This has the effect of confusing the semantic model when looking for symbol information for SyntaxNode: because the appearance of SyntaxNode in the RHS of the BoundConditionalAccessExpression is lower in the bound tree than the appearance of SyntaxNode in the LHS, the BoundConditionalReceiver associated with the RHS will be used to lookup symbols. The primary effect of this is a failure to return the actual method called in situations such as the following:

class C
{
    public int I;

    public static C M(long l) { return null; }
    public static C M(int i) { return null; }

    public static int Main()
    {
        return (int)(M(0)?.I);
    }
}

If semantic information is requested for "M" in the call to "M(0)", the returned information will contain both overloads of "M" instead of the actual overload chosen.

The actual fix is simple:
- Mark all BoundConditionalReceiver nodes as compiler generated, thus eliminating them from consideration for MemberSemanticModel.GetBoundNodes
- Stop digging through the syntax tree for conditional expressions when getting the bindable parent node of an expression

Unfortunately, the recommendation service took a dependency on seeing the BoundConditionalReceiver in the tree when requesting type information for the bound LHS of a member binding expression via SemanticModel.GetTypeInfo: if the type of the original receiver for the conditional access is T?, the type of the BoundConditionalReceiver is T. Instead of relying on this behavior, the recommendation service has been updated to perform this transformation manually.
***NO_CI***
 (changeset 1386679)

There are some warnings that we havent implemented in Roslyn but have kept around because someone could have done a /nowarn and we would have complained in the past if we didnt recognize that warning number.

However now that /nowarn doesnt complain about unrecognized warnings (because it needs to support userdiags) there's no point keeping them. (changeset 1386659)

VB Unstructured Exception Handling: Finish work to comply with requirement that IL shall only differ between Release and Debug. Enable/fix all Unstructured Exception Handling unit-tests.
***NO_CI***
 (changeset 1386638)