================
@@ -1198,16 +1204,33 @@ bool createParameters(NewFunction &ExtractedFunc,
}
}
+ // A reference-to-array parameter's declared type always spells out
+ // every dimension's bound, including the outermost one (unlike a
+ // decayed C pointer, which drops it). For a variable-length array,
+ // that bound is an arbitrary expression -- typically another
+ // variable's name -- that isn't in scope in the extracted
+ // function's declaration, and even making it an explicit preceding
+ // parameter doesn't help: the two don't line up at the call site
+ // (the callee's declared bound and the caller's concrete array type
+ // don't reference-bind just because both happen to be named the
+ // same). So this is rejected unconditionally, in both languages --
+ // C still gets a chance to extract it below, by decaying to a
+ // pointer instead of keeping a reference.
+ if (Kind == ParamPassKind::Reference && TypeInfo->isArrayType() &&
+ TypeInfo->isVariablyModifiedType() && LangOpts.CPlusPlus)
----------------
ArcsinX wrote:
Maybe we can also add a check for pointers to VLA.
E.g. this case produces uncompilable code
```cpp
void foo(int N) {
int (*A)[N];
[[ (*A)[0] = 1; ]]
}
```
https://github.com/llvm/llvm-project/pull/230046
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
