Error(s) with code-gen of if-statements

[cduck]

Bug report by https://github.com/randyshee/yquantum/blob/main/2025/team_solutions/charlie_and_delta/bug_report.ipynb

There are several related errors with the QASM2 if-statements that are emitted by Bloqade that cause parse errors for Qiskit.

If-statement with one basic operation in it works:

from bloqade import qasm2
from bloqade.qasm2.emit import QASM2 # the QASM2 target
from bloqade.qasm2.parse import pprint # the QASM2 pretty printer
from bloqade.qasm2.parse import pprint, spprint, Console
from qiskit import QuantumCircuit

circ_nested = None
del circ_nested

@qasm2.extended
def circ_nested():
    qr = qasm2.qreg(4)
    cr = qasm2.creg(4)

    qasm2.measure(qr, cr)
    
    if cr == 0:
        # Different errors depending on the code in this block:
        qasm2.x(qr[3])

    return qr

target = QASM2()
ast = target.emit(circ_nested)
qasm = spprint(ast, console=Console(no_color=True, force_jupyter=False, force_interactive=False, force_terminal=False))
print(qasm)
circuit = QuantumCircuit.from_qasm_str(qasm)

Two doesn't work:

    if cr == 0:
        # Different errors depending on the code in this block:
        qasm2.x(qr[3])
        qasm2.x(qr[3])

OPENQASM 2.0;
include "qelib1.inc";
qreg qr[4];
creg cr[4];
measure qr -> cr;
if (cr == 0) {
  x qr[3];
  x qr[3];
}
-----
QASM2ParseError: '<input>:6,13: needed a gate application, measurement or reset, but instead saw {'

The error message makes it clear that the if-statement body is restricted.

[weinbe58]

@Roger-luo I think there are two related issues here:

1. qasm2.main We should probably restrict this when lowering the python kernel
2. qasm2.extended needs to error when trying to emit qasm2
3. We can implement some other passes from Py2QASM to help transform the extended dialect into something more compatible with QASM2

[david-pl]

For reference, from the QASM2 specification:

The if statement conditionally executes a quantum operation based on the value
of a classical register. This allows measurement outcomes to determine future quantum
operations. We choose to have one decision register for simplicity. This register is interpreted
as an integer, using the bit at index zero as the low order bit. The quantum operation
executes only if the register has the given integer value. Only quantum operations, i.e. built-
in gates, gate (and opaque) subroutines, preparation, and measurement, can be prefaced
by if. A quantum program with a parameter that depends on values that are known only
at run-time can be rewritten using a sequence of if statements. Specifically, for a single-
parameter gate with n bits of precision, we may choose to write 2n statements, only one
of which is executed, or we can decompose the parameterized gate into a sequence of n
conditional gates.

TL;DR: only single-line if statements are supported.

[david-pl]

@weinbe58 I thought a bit more about your idea for a custom if statement. So how about, as a first step, we add a custom Qasm2If statement, that verifys the length of the then block (and possible other things, such as whether the operations in the block are valid for qasm2 conditionals). That would give us a compiler error for now.
Later, we can think about adding rewrite passes for qasm2.extended (your 3.).

Would that make sense?

[Roger-luo]

yeah we kinda extended the parser a bit to let if block accept multiple statement (this is really annoying when working with QASM2). So I guess for compatibility reasons we can remove this syntax now.

Then we could also remove the QASM2 codegen to reject scf.if that has more than one statement in the body. You don't need a special QASM2If because this is still scf.If except you error on multiple statement in the body.

For validation, I think we can do this in a separate pass instead of doing it within the statement verify - I think creating a new statement just for this is a bit too restrictive when a simple check is possible. We don't need to worry about validation in this issue because we haven't implemented any formal validation in this package yet.

[weinbe58]

Do we want to add a rewrite that breaks the multi-line body into multiple if statements?

[Roger-luo]

That could also work without having to ban this so yes otherwise this is getting too restrictive