Improve error message for mismatching unit ids when adding conditional gates

[CalMacCQ]

The following method works fine for adding a condtional-X gate to a Circuit.

from pytket import Circuit

circ = Circuit()
q = circ.add_q_register("q", 1) # default "q" register
a = circ.add_c_register("a", 1)

circ.X(q[0], condition_bits=[a[0]], condition_value=1)

Notice how I am passing the Qubit q[0] into the Circuit.add_gate method.

However if I replace q[0] with 0....

from pytket import Circuit

circ = Circuit()
q = circ.add_q_register("q", 1) # default "q" register
a = circ.add_c_register("a", 1)

circ.X(0, condition_bits=[a[0]], condition_value=1)

I get the following RuntimeError

--
RuntimeError                              Traceback (most recent call last)
Cell In[16], line 1
----> 1 circ.X(0, condition_bits=[a[0]], condition_value=1)

RuntimeError: Unable to cast Python instance of type <class 'list'> to C++ type 'std::__1::vector<unsigned int, std::__1::allocator<unsigned int>>'

Its not at all obvoius from this error what has gone wrong here. This error message should make the fix clearer.

In fact, arguably this shouldn't give an error in pytket as in the current interface we support referencing qubits with Qubits and integers. However not sure how practical overload merging would be.

[github-actions]

This issue has been automatically marked as stale.

[willsimmons1465]

This error is thrown by pybind.
When we expose multiple methods with the same name and arity, pybind creates one python method with that name and performs type inference on the arguments to determine which of the underlying methods to call. It spits out this generic error message whenever this type inference fails.

In order for us to change the error message, we would have to take manual control away from pybind for handling this. I messed around last year with only ever exposing e.g. _Circuit, and defining the python class Circuit which acts as a thin wrapper around it and I manually do the type inference. This has the added benefits of being able to fully customise the docstring and type annotations as well (which can be a massive improvements to the API docs to cut down on all of the pointless "Overloaded function" blocks with up to 12 subtly different cases for essentially the same method), as well as making everything in pytket searchable through the python source. However, doing this on mass is a very large amount of work and increases the complexity of the codebase by adding yet another distinct binding layer to maintain.

pybind11 does offer the option of doing the type inference in the C++ binders, so it might be possible to do the same "taking over the inference role to only expose a single method" trick without adding a new layer for us to maintain. Still a lot of work, and we might not have fine-grained control over the type annotations in the docs/mypy (though we would still benefit from complete control over docstrings which would help), but at least this is something that can be done much more incrementally. Maybe we can try building an example of this and assess how scalable it is. I'll have a crack at that this week and report back.

[CalMacCQ]

Thanks Will

Here's another example where the error message is not so clear

from pytket import Bit, Qubit
from pytket.circuit import CircBox, Circuit
from pytket.circuit.logic_exp import reg_eq

box = CircBox(Circuit(2, name="C").H(0).CX(0, 1))
outer_circ = Circuit(4, 2).H(0).CX(0, 1).Measure(0, 0).Measure(1, 1)
c_reg = outer_circ.get_c_register("c")

outer_circ.add_circbox(circbox=box,
                       args=[Qubit(2), Qubit(3)],
                       condition_bits=[Bit(0), Bit(1)],
                       condition_value=reg_eq(c_reg, 1))

Here's the error

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[16], line 1
----> 1 outer_circ.add_circbox(circbox=box, args=[Qubit(2), Qubit(3)], condition_bits=[Bit(0), Bit(1)], condition_value=reg_eq(c_reg, 1))

RuntimeError: Unable to cast Python instance of type <class 'pytket.circuit.logic_exp.RegEq'> to C++ type 'unsigned int'

In this case the fix is to use the condition=reg_eq(c_reg, 1)) arg as a replacement for condition_bits and condition_value.