This change also ensures that we get non-fatal Watson reports when we're in the continuation of a call to get generic type arguments or argument values from the debugee process.

Fixes #309 (Incremental build fix)

Fix #295

Review feedback from Jason

Program for running tests in parallel

These suites are failing on a number of machines without this change.
Disabling.

fix test break I introduced.

Don't preselect enums on backspace

When backspacing in VB, we used to prefer preselected items over text
matches. Dev12 does this for object creation preseletion, but not enum
preselection. We should match that behavior.

Fixes #287.

I added encoding check in code that are shared by multiple tests.

this change make encoding check to be explicitly requested

Small updates to the change

- Sort the output by ellapsed time
- Disable a suite which was failing

Allow 'is' for static classes for compatibility. Fixes #324.

Currently tests are run on the developers machine using

> msbuild BuildAndTest.proj

This takes ~25 minutes to complete on the CI machine and ~45 minutes on
a developer machine (developer machines run a greater number of suites).
This is not conductive to quick bug fixes.

Ideally the infrastructure would switch over to using xunit 2 which can
handle parallelizing the tests for us.  Unfortunately this isn't an
option because:

1. The 32 bit runner quickly OOMS on our suites
2. A number of our tests can't run in 64 bit

Long term xunit2 is still our goal.  Short term though this tool will
use process level parallelization to run our suites.

Update the DeclarePublicAPI analyzer to consider return types

Manage the lifetime of DkmClrValue objects

Move VB end construct code fix away from CodeGenerator

We have two targets that each modify a compiled binary in-place:
ApplyOptimizations and FakeSign. They each write out their own sentinel
file to help keep incremental build healthy and quick. However, this
causes problems when you need to run both (e.g. on a developer box
compiling Release binaries).
This change moves the sentinel file into a third task that depends on
the other two. I hope it's obvious how to extend this if more "post
compile binary modification" targets are added later.

We need to attach the same information to DkmEvaluationResults and
DkmEvaluationResultEnumContexts, but they have different lifetimes.  Enum
contexts, which are, in some sense, the children of evaluation results are
closed first.  As such we don't want to clean up our attached information
until the evaluation result is closed.  When we are informed that this has
happened (by a call to EvalResultDataItem.OnClose), we call Close on the
associated DkmClrValue.

We hope that actively managing the lifetime of DkmClrValue instances will
improve perf by reducing the amount of work that the debugger has to do on
each evaluation (which is proportional to the number of live
DkmClrValues).

Unfortunately, this will not improve perf in the large-array case because
DkmEvaluationResults are not closed as they are scrolled out of view.

Associated with DevDiv #1094428 (not yet on GitHub).

The DeclarePublicAPI draws attention to changes to the public API
surface area in order to prevent breaking changes and to encourage
careful design when adding new APIs. It requires that every project with
a public API capture the public types and members in a file called
PublicAPI.txt, and creates errors when this file does not match the
actual symbols in the project. This makes changes very obvious in code
reviews and source control history.

Currently PublicAPI.txt does not capture the return types of members
(field, properties, methods, and events) so these can changes without it
being caught by the tool. This changes updates the analyzer to include
the return type.

A nice feature of the associated code fix is that it sorts PublicAPI.txt
alphabetically, and since type names are fully qualified this naturally
groups together types in the same namespace, members in the same type,
overloads, etc. To preserve this property we put the return type at the
end of the line, after the type and member name.

while creating pull request, I left out this change

but continue to report the diagnostic if "strict" mode is requested.
Fixes #324

fix vsct file to not have dupliate button definitions

Add MetadataVisualizer project to Roslyn solutions

Construct EnumContextDataItems on demand.  Also, don't access their properties before checking for null.

disable some tests due to interface tests.

these tests will be re-enabled once we patch build machine with new interfaces.

Conflicts:
	src/Roslyn.sln

Replace nonbreaking space (0xA0) with a regular space.

We need to attach the same information to DkmEvaluationResults and
DkmEvaluationResultEnumContexts, but they have different lifetimes.  Enum
contexts, which are, in some sense, the children of evaluation results are
closed first.  As such we don't want to clean up our attached information
until the evaluation result is closed.  When we are informed that this has
happened (by a call to EvalResultDataItem.OnClose), we call Close on the
associated DkmClrValue.

We hope that actively managing the lifetime of DkmClrValue instances will
improve perf by reducing the amount of work that the debugger has to do on
each evaluation (which is proportional to the number of live
DkmClrValues).

Unfortunately, this will not improve perf in the large-array case because
DkmEvaluationResults are not closed as they are scrolled out of view.

Associated with DevDiv #1094428 (not yet on GitHub).

It is likely the character was inserted by accident. There is no reason
to use it in the code. It is a valid Unicode whitespace, but not a valid
ASCII. Since the containing file has no encoding bytemarks, the behavior
of this char is dependent on default encoding. Indeed, the file compiles
on Windows and fails on Linux.
As far as I could find, this is the only case where this character is
used in actual code (not in tests) in Roslyn codebase.