馃 AI text below 馃
Problem statement
The parser has three related gaps in how it validates register references and comparison values:
-
Well-formed identifier checking. Neither validateTargets (for qubit targets in gate calls) nor parseBitRegisterRef (for classical bits in if conditions) rejects malformed register names on lexical grounds. Inputs like c2] (dangling ]) or @# (not an identifier at all) slip through the shape check. On the qubit side the eventual definedRegisters lookup rejects them indirectly with an "unknown register" error; on the classical side there is no lookup, so the error surfaces even later in the DD backend.
-
Classical bit reference against declared registers. validateTargets verifies that qubit targets exist in definedRegisters and that the bit index fits within the register size. Nothing analogous runs for the register mentioned in an if condition: if (nonexistent == 1) and if (c[999] == 1) (with creg c[3]) both parse without complaint.
-
expectedValue within the register's representable range. if (c == 999) on creg c[3] (3 bits, values 0 to 7) is impossible to satisfy but the parser accepts it. The comparison always evaluates to false at runtime.
Surfaced during the review of PR #463.
Proposed solution
Add the three missing checks and share them where the same code path applies to qubits and classical bits:
- Extract a small
isValidIdentifier helper (letter or underscore followed by letters, digits, or underscores). Use it in validateTargets (before the definedRegisters lookup) and in parseBitRegisterRef (before returning the reference).
- Extend
preprocessCode so it also runs an existence + bounds check on the register mentioned in the condition of an if, using the same definedRegisters map already available there. The exact placement (inside parseClassicConditionExpression or a separate validateClassicCondition pass) is an implementation detail.
- Reject
expectedValue values that exceed what the referenced register can hold (whole register or single bit).
Out of scope
- Reserved-word checking (
qreg, creg, measure, etc.). Separate semantic concern.
- Compound expressions in the condition (out of scope of the parser as a whole today).
馃 AI text below 馃
Problem statement
The parser has three related gaps in how it validates register references and comparison values:
Well-formed identifier checking. Neither
validateTargets(for qubit targets in gate calls) norparseBitRegisterRef(for classical bits inifconditions) rejects malformed register names on lexical grounds. Inputs likec2](dangling]) or@#(not an identifier at all) slip through the shape check. On the qubit side the eventualdefinedRegisterslookup rejects them indirectly with an "unknown register" error; on the classical side there is no lookup, so the error surfaces even later in the DD backend.Classical bit reference against declared registers.
validateTargetsverifies that qubit targets exist indefinedRegistersand that the bit index fits within the register size. Nothing analogous runs for the register mentioned in anifcondition:if (nonexistent == 1)andif (c[999] == 1)(withcreg c[3]) both parse without complaint.expectedValuewithin the register's representable range.if (c == 999)oncreg c[3](3 bits, values 0 to 7) is impossible to satisfy but the parser accepts it. The comparison always evaluates to false at runtime.Surfaced during the review of PR #463.
Proposed solution
Add the three missing checks and share them where the same code path applies to qubits and classical bits:
isValidIdentifierhelper (letter or underscore followed by letters, digits, or underscores). Use it invalidateTargets(before thedefinedRegisterslookup) and inparseBitRegisterRef(before returning the reference).preprocessCodeso it also runs an existence + bounds check on the register mentioned in the condition of anif, using the samedefinedRegistersmap already available there. The exact placement (insideparseClassicConditionExpressionor a separatevalidateClassicConditionpass) is an implementation detail.expectedValuevalues that exceed what the referenced register can hold (whole register or single bit).Out of scope
qreg,creg,measure, etc.). Separate semantic concern.