The LDM has decided that the following rules should be in effect when
calculating definite assignment for local functions:

  1. The entry point to a local function is always reachable.
  2. Variables captured in local functions are definitely assigned if
  they are definitely assigned in all branches into the local function.

It turns out these rules were almost exactly what the compiler already
implemented, but there was a bug in captured variable detection that
meant that variables captured in lambdas within local functions were
sometimes not counted as captured. This change fixes the bug around
capturing, which should cause the compiler to conform to this
specification.

Fixes #17829

The argument that stands for the receiver of a `ref` extension method must be always passed as `ref`

* Make sure 'in' is honored in cref parsers/completion lists

* Clean up

* PR Comments

* Document API breaking change

This looks like simple oversight. The binder being used for type
parameters was retrieved from the binder factory since we didn't
explicitly pass one. For members this is correct since the binder
factory can be queried for top-level binders. This is not correct for
local functions because they use method body binders, which are not
accessible from the binder factory.

Mostly, this doesn't matter. The exception is when you try to
speculatively bind an attribute on a type parameter. Here, you need an
in-method binder and VS will crash if it's the wrong binder. This was
uncommon since attributes on type parameters are not permitted in local
functions, but VS should not crash.

Fixes #17814 

Merging on behalf of @zaytsev-victor. Thanks for the contribution!

* Arrays/pointers/nullables of intrinsic types are intrinsic. Type of array creations is apparent. Handle out vars.

The argument that stands for the receiver of a `ref` extension method must be always passed as `ref`

Snap master into dev15.6.x

Skip pseudo-variables with use-site errors

* Simplify codegen for struct declaration patterns
* Move box to perform the conversion only when the test succeeds

Avoid skewing analysis time when logging analyzer times

Be resilient to the OOP process going away in TodoComments and DesignerAttribute processing

Fix ArgumentException when creating diagnostic with effective severit…

* Reject ref as an invalid modifier for accessors

* Updated test plan

* PR Comments

* Accept ref in parsing, and error out in binding

* Update test plan

* Fix build break

Do not autodisable verification based on unsafe compilation option.

MemberSemanticModel.GetBoundLambdaOrQuery should use GetAdjustedNodePosition helper to ensure proper handling of zero-length tokens at the beginning of a syntax node. (#24049)

Fixes #23883.

Fixes VSO 547049

* added Microsoft.VisualStudio.IntegrationTest.Setup.vsix to signing

* removed duplicated entry.

* now test setup vsix is part of signtooldata.json no need to have special case for it.

also made exception to have better message

Merging on behalf of @alrz. Thanks!

There were two separate bugs here:
    1) Fixing up the MVID of the binary was being done *after* signing,
    when it should be done before.

    2) System.Reflection.Metadata has a bug where calculating the
    checksum occurs before signing, when it should happen after.

Both of these isues have been fixed and the helper utility that checks
to see whether a binary was signed has been changed to verify that the
signature and checksums of the binaries are valid, not just that the
signed bit is set on the assembly.

Fixes #23521

Merge dev15.6-preview2 to master

Merge remote-tracking branch 'dotnet/dev15.6-preview2' into merges/dev15.6-preview2-to-master-20180104-080013

Removing the rename command target from Roslyn as it will be part of core editor platform