Fix digit escaping in OpenQASM 3 identifier

qiskit/qasm3/exporter.py

@@ -174,12 +174,17 @@ class DefcalInstruction:
     }
 )
 
-# This probably isn't precisely the same as the OQ3 spec, but we'd need an extra dependency to fully
-# handle all Unicode character classes, and this should be close enough for users who aren't
-# actively _trying_ to break us (fingers crossed).
-_VALID_DECLARABLE_IDENTIFIER = re.compile(r"([\w][\w\d]*)", flags=re.U)
-_VALID_HARDWARE_QUBIT = re.compile(r"\$[\d]+", flags=re.U)
-_BAD_IDENTIFIER_CHARACTERS = re.compile(r"[^\w\d]", flags=re.U)
+# This is deliberately more restrictive than the OQ3 spec - the builtin `re` module has weak Unicode
+# support, and this need here doesn't rise to the level of adding the third-party `regex` as a
+# dependency.  Python's `\w` matches way too much (basically anything in Unicode classes [L?] and
+# [N?], plus _), while `\d` matches way too little (only [Nd]) to be used as a negation.  As a
+# compromise, we allow ASCII letters, Greek letters (since they're a small, contiguous block in
+# Unicode that can be specified easily, and physicists like them), _ and [0-9].  Everything else is
+# escaped.
+_ALPHA = r"a-zA-Z\u0370-\u03ff"  # ASCII alpha, plus the "Greek and Coptic" Unicode block.
+_VALID_DECLARABLE_IDENTIFIER = re.compile(rf"[{_ALPHA}_][{_ALPHA}_0-9]*", flags=re.U)
+_VALID_HARDWARE_QUBIT = re.compile(r"\$[0-9]+", flags=re.U)
+_BAD_IDENTIFIER_CHARACTERS = re.compile(rf"[^{_ALPHA}_0-9]", flags=re.U)
 
 
 class Exporter:

releasenotes/notes/qasm3-fix-identifiers-c2adba914620b486.yaml

@@ -0,0 +1,9 @@
+---
+fixes:
+  - |
+    The OpenQASM 3 exporter (:func:`.qasm3.dumps` and :func:`~.qasm3.dump`) will now correctly
+    escape Unicode "number" characters other than ASCII digits, when used in identifiers.  Previously,
+    characters like superscripts would remain in the output, which is not valid OpenQASM 3.
+  - |
+    The OpenQASM 3 exporter (:func:`.qasm3.dumps` and :func:`~.qasm3.dump`) will now correctly
+    escape identifiers whose name begins with an ASCII digit.

test/python/qasm3/test_export.py

@@ -113,8 +113,11 @@ def setUpClass(cls):
         # be useful for the tests must _never_ have false positive matches.  We use an explicit
         # space (`\s`) or semicolon rather than the end-of-word `\b` because we want to ensure that
         # the exporter isn't putting out invalid characters as part of the identifiers.
+        alpha = r"a-zA-Z\u0370-\u03ff"
+        id_first = rf"[{alpha}_]"
+        id_cont = rf"[{alpha}_0-9]"
         cls.register_regex = re.compile(
-            r"^\s*(let|(qu)?bit(\[\d+\])?)\s+(?P<name>\w+)[\s;]", re.U | re.M
+            rf"^\s*(let|(qu)?bit(\[\d+\])?)\s+(?P<name>{id_first}{id_cont}*)[\s;]", re.U | re.M
         )
         scalar_type_names = {
             "angle",
@@ -127,7 +130,7 @@ def setUpClass(cls):
         cls.scalar_parameter_regex = re.compile(
             r"^\s*((input|output|const)\s+)?"  # Modifier
             rf"({'|'.join(scalar_type_names)})\s*(\[[^\]]+\])?\s+"  # Type name and designator
-            r"(?P<name>\w+)[\s;]",  # Parameter name
+            rf"(?P<name>{id_first}{id_cont}*)[\s;]",  # Parameter name
             re.U | re.M,
         )
         super().setUpClass()
@@ -1477,21 +1480,27 @@ def test_registers_have_escaped_names(self):
         """Test that both types of register are emitted with safely escaped names if they begin with
         invalid names. Regression test of gh-9658."""
         qc = QuantumCircuit(
-            QuantumRegister(2, name="q_{reg}"), ClassicalRegister(2, name="c_{reg}")
-        )
-        qc.measure([0, 1], [0, 1])
+            QuantumRegister(2, name="q_{reg}"),
+            ClassicalRegister(2, name="c_{reg}"),
+            QuantumRegister(2, name="²"),
+            ClassicalRegister(2, name="2c"),
+            QuantumRegister(2, name="abc?!abc$%^&"),
+            ClassicalRegister(2, name="?!abc$%^&"),
+        )
+        qc.measure(qc.qubits, qc.clbits)
         out_qasm = dumps(qc)
         matches = {match_["name"] for match_ in self.register_regex.finditer(out_qasm)}
-        self.assertEqual(len(matches), 2, msg=f"Observed OQ3 output:\n{out_qasm}")
+        self.assertEqual(len(matches), 6, msg=f"Observed OQ3 output:\n{out_qasm}")
 
     def test_parameters_have_escaped_names(self):
         """Test that parameters are emitted with safely escaped names if they begin with invalid
         names. Regression test of gh-9658."""
         qc = QuantumCircuit(1)
-        qc.u(Parameter("p_{0}"), 2 * Parameter("p_?0!"), 0, 0)
+        qc.u(Parameter("p_{0}"), 2 * Parameter("2p"), Parameter("a²"), 0)
+        qc.rz(Parameter("!$abc%$&"), 0)
         out_qasm = dumps(qc)
         matches = {match_["name"] for match_ in self.scalar_parameter_regex.finditer(out_qasm)}
-        self.assertEqual(len(matches), 2, msg=f"Observed OQ3 output:\n{out_qasm}")
+        self.assertEqual(len(matches), 4, msg=f"Observed OQ3 output:\n{out_qasm}")
 
     def test_parameter_expression_after_naming_escape(self):
         """Test that :class:`.Parameter` instances are correctly renamed when they are used with