================
@@ -6544,6 +6544,14 @@ bool InterpretOffsetOf(InterpState &S, CodePtr OpPC,
const OffsetOfExpr *E,
// When generating bytecode, we put all the index expressions as Sint64
on
// the stack.
int64_t Index = ArrayIndices[ArrayIndex];
+ // Reject negative indices and unsigned indices that wrapped to negative
+ // after the Uint64->Sint64 cast (e.g. __uint128_t >=
0x8000000000000000).
+ if (Index < 0) {
----------------
marlus wrote:
Thanks. The Index < 0 guard here doesn't cover values that overflow 64 bits
entirely — for example, __uint128_t(2^64) gets silently truncated to 0 when
cast through PT_Uint64 in Compiler.cpp, so Index arrives as 0 and bypasses the
guard.
Fixed in Compiler.cpp: AP types (PT_IntAP / PT_IntAPS, i.e. __uint128_t,
__int128, etc.) now cause constant evaluation to fail before any bytecode is
emitted, avoiding the truncation entirely. Added a test for ((__uint128_t)1 <<
64) to cover this case explicitly.
https://github.com/llvm/llvm-project/pull/204139
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
