Possible fix for issue #51

[danielcweber]

According to the Roslyn sources (here and here) the initial state of a state machine that Roslyn creates for an async method is -1. Code Contracts assumes that for Roslyn, the initial state is 0 and thus fails in rewriting async methods compiled by Roslyn. See #51 for more details.

[matteo-mosca]

I'm actually looking forward to this. This is the only thing preventing me to migrate my current work to .net 4.6 and aspnet v5.

[Daniel-Svensson]

👍 This solves my problems with async.and roslyn.

And it also seems to resolve #38

[SergeyTeplyakov]

Few questions/suggestions:

1. Please update a comment and put following link as a proof that Roslyn is using the same constants as old compiler: https://github.com/dotnet/roslyn/blob/b6484300dfafb43af0c27e542ec457a7583e1aa8/src/Compilers/CSharp/Portable/Lowering/StateMachineRewriter/StateMachineStates.cs

2. Is there any changes in tests? What the impact from end-user perspective?

3. I see another place in the code with a similar logic:

Extractor.cs:GetContractClumpFromMoveNext

var initialState = moveNext.IsAsync && !isRoslyn ? -1 : 0;
moveNext.MoveNextStartState = initialState;
originalContractPosition = null;
int statementIndex;
Contract.Assume(moveNext.Body != null);
Contract.Assume(moveNext.Body.Statements != null);
int blockIndex = ContractStartInMoveNext(this.contractNodes, moveNext, out statementIndex, iteratorMethod, isRoslyn);

I was unable to find usages for MoveNextStartState but for the sake of consistency I would suggest to change this as well.

Thanks a lot for investing time in this project!

[SergeyTeplyakov]

Closed as dup.