================
@@ -3324,6 +3359,20 @@ llvm::Constant
*CodeGenFunction::EmitCheckTypeDescriptor(QualType T) {
DiagnosticsEngine::ak_qualtype, (intptr_t)T.getAsOpaquePtr(),
StringRef(),
StringRef(), std::nullopt, Buffer, std::nullopt);
+ if (IsBitInt) {
+ // The Structure is: 0 to end the string, 32 bit unsigned integer in target
+ // endianness, zero.
+ char S[6] = {'\0', '\0', '\0', '\0', '\0', '\0'};
----------------
earnol wrote:
This was initial idea proposed in
https://github.com/llvm/llvm-project/issues/64100.
Parsing from the string is a reasonable idea, but
1. It will put additional performance burden on the runtime as we will need to
do parsing in the runtime. It will require either regexp or FSM for parsing.
Please read comments above when i have been requested to change CHECK to DCHECK
for the performance reasons. And using sscanf or regexp will definitely put
much serious burden on runtime compared to icmp.
2. This string is not used for anything but for diagnostics currently. Probably
putting additional meaning for it is fine, but ... (see next point)
3. It will go against already implemented principles where we using pre-parsed
data. If we are allowed to parse the type we do not need Type and Kind members
of the TypeDescriptor. All this information can be extracted from the type
string as it will become the single source of truth.
https://github.com/llvm/llvm-project/pull/96240
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
