I am writing C# unit tests using NUnit and NSubstitute. I am testing a class which will attempt to retrieve objects from a config provider implementing the following interface:
public interface IConfigProvider<T> {
T GetConfig(int id);
T GetConfig(string id);
}
The class being tested only uses the int version of GetConfig
so in the SetUpFixture I do the following to set up a mocked config provider that will always return the same dummy object:
IConfigProvider<ConfigType> configProvider = Substitute.For<IConfigProvider<ConfigType>>();
configProvider.GetConfig(Arg.Any<int>()).Returns<ConfigType>(new ConfigType(/* args */);
This runs absolutely fine if that TestFixture is the only one being run. However, in a different TestFixture in the same assembly, I check for received calls like this:
connection.Received(1).SetCallbacks(Arg.Any<Action<Message>>(), Arg.Any<Action<long>>(), Arg.Any<Action<long, Exception>>());
If these Received
tests run before the config provider tests, then the config tests fail in the SetUpFixture with an AmbiguousArgumentsException:
Here.Be.Namespace.ProfileManagerTests+Setup (TestFixtureSetUp):
SetUp : NSubstitute.Exceptions.AmbiguousArgumentsException : Cannot determine argument specifications to use.
Please use specifications for all arguments of the same type.
at NSubstitute.Core.Arguments.NonParamsArgumentSpecificationFactory.Create(Object argument, IParameterInfo parameterInfo, ISuppliedArgumentSpecifications suppliedArgumentSpecifications)
at System.Linq.Enumerable.<SelectIterator>d__7`2.MoveNext()
at System.Collections.Generic.List`1..ctor(IEnumerable`1 collection)
at NSubstitute.Core.Arguments.MixedArgumentSpecificationsFactory.Create(IList`1 argumentSpecs, Object[] arguments, IParameterInfo[] parameterInfos)
at NSubstitute.Core.Arguments.ArgumentSpecificationsFactory.Create(IList`1 argumentSpecs, Object[] arguments, IParameterInfo[] parameterInfos, MatchArgs matchArgs)
at NSubstitute.Core.CallSpecificationFactory.CreateFrom(ICall call, MatchArgs matchArgs)
at NSubstitute.Routing.Handlers.RecordCallSpecificationHandler.Handle(ICall call)
at System.Linq.Enumerable.WhereSelectArrayIterator`2.MoveNext()
at System.Linq.Enumerable.FirstOrDefault[TSource](IEnumerable`1 source, Func`2 predicate)
at NSubstitute.Routing.Route.Handle(ICall call)
at NSubstitute.Proxies.CastleDynamicProxy.CastleForwardingInterceptor.Intercept(IInvocation invocation)
at Castle.DynamicProxy.AbstractInvocation.Proceed()
at Castle.Proxies.IConfigProvider`1Proxy.GetConfig(Int32 id)
at Here.Be.Namespace.ProfileManagerTests.Setup.DoSetup()
What's really confusing me is that I can observe this effect even between test runs - if I use the NUnit GUI to run the Received
tests alone, and then run the config tests alone, the config tests will fail. If I then immediately run the config tests again, they will pass.
Things I've tried:
configProvider.GetConfig(Arg.Any<string>()).Returns...
as well, in case the overloading was the problem.As it happens, the tests I'm using will only ever call the GetConfig
method with values of 0 or 1, so I can just provide Returns
specifications for those two values and not use matching at all, but I want to understand how to fix this more generally.
I had similar errors that started when I switched Microsoft test runner to VSTest.Console
(they did not happened when running under MSTest.exe
).
As it was advised in David's answer, the errors were caused by calls to non-substituted methods with Arg.*
parameters. Arg.Any
were passed to actual code methods, that where called without Returns
or Received
relevant methods.
To scan my test library for such issues I used search with regular expression to find rows with Arg.
but not Arg.
following by Returns
or preceded by Received
(?=^.*Arg.*$)(?=^((?!Arg.*\.Returns).)*$)^((?!\.Received\(.*Arg.).)*$
It's not bullet proof filter (e.g it doesn't exclude multi-line statements), but it helps to reduce number of calls to check.
Ambiguous arguments is when NSubstitute compares the arguments to the call it is currently working with, to the stack of "argument matchers" it has (each time Arg.Blah
is called, an the argument matcher is added to that stack), and it is unable to resolve which argument goes where.
Normally this is caused by having a call like blah(null, null)
, with a single argument matcher queued up, but it can also be caused by the stack getting out-of-sync due to an arg matcher being used outside of call configuration, or as an argument to a non-virtual method.
Version 1.8.0 (released after your question) includes slightly improved detection of the latter case, so that may be worth trying.
Other than that, I've had this problem a few times and have used the following (painful) approach.
Arg.xyz
that could queue up an argument matcher in either test. Make sure it is used as part of a call configuration. Sometimes working out which call is problematic can be done by commenting out lines or replacing arg matchers with other values.Sometimes the problem may be due to a previous fixture, so you may need to workout the previous fixture and explore there as well. :(
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With