================
@@ -170,7 +170,7 @@ class TransferVisitor : public
ConstStmtVisitor<TransferVisitor> {
auto *RHSVal = Env.getValue(*RHS);
if (RHSVal == nullptr)
- break;
+ RHSVal = Env.createValue(LHS->getType());
----------------
ymand wrote:
> I haven't quite figured out what exactly does a null value represents. Is it,
> like, "unimplemented"? If yes, creating a new value is probably better.
Sorry -- we really should have a dev guide for Clang Dataflow. Yes, it
represents "unmodeled". It's an incomplete part of the system, which can have
assorted causes. Our general approach is to on-demand create values in these
cases (that is, wait until a point where have a fresh/blank value will be
useful), but I don't believe we have a hard and fast rule. But, as your test
demonstrates, the current behavior is definitely wrong.
I'm somewhat inclined towards your choice, because it means that we'll have
consistent reasoning is this simple case:
```
Foo = unknown();
int FooAtB = Foo;
if (cond()) {
// something involving FooAtB
} else {
// something else involving FooAtB
}
```
With your current approach, a fresh value is populated and the branches will
consider the same fresh value. otherwise, anything that on-demand introduces
the value used by FooAtB, will result in separate copies for the two branches.
https://github.com/llvm/llvm-project/pull/178943
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
