DumpOperation doesn't need its argument to have an adjoint

[tcNickolas]

DumpOperation uses only op, not its adjoint, so it should work fine with operations that do not have an adjoint. I ran into a case when I need to print the matrix of a unitary operation that is implemented using mutable variables (so not automatically or easily adjointable) but includes a sequence of gates that has a matrix representation. And, looking at the SingleQ example in https://learn.microsoft.com/en-us/azure/quantum/testing-debugging, dump_operation doesn't have the same restriction.

[swernli]

Hi @tcNickolas! Yes, this one was debated for a while on the team. Part of my hesitation to relax the restriction was a worry about operations that included measurement of the input qubits, since that could break the logic used to calculate the effective unitary matrix from the state of the entangled qubits. Of course, someone could defeat this restriction by including an explicit adjoint implementation that is a no-op, so it's a weak restriction at best.

Out of curiosity, if an operation that included measurement was passed to DumpOperation, what would be your expectation? Should it fail to compile, fail at runtime, or simply display the result of the calculation as a best effort?

[tcNickolas]

An operation that includes measurements can implement a unitary (using RUS), and in the old QDK DumpOperation used to just display the result of whatever happens, so I think that's what it makes sense to do. If somebody passes a non-unitary to DumpOperation, well, they know the risks they're taking :-)