[clang] [C23] Select the correct promoted type for a bit-field (PR #89254)

Timm Baeder via cfe-commits cfe-commits at lists.llvm.org
Thu Apr 18 22:13:22 PDT 2024


tbaederr wrote:

> @tbaederr -- this triggers an assertion with the new constexpr interpreter in `clang/test/AST/Interp/intap.cpp` but I'm not certain I understand why. Can you help me figure out how to fix that?
> 
> https://buildkite.com/llvm-project/github-pull-requests/builds/56785#018ef1ee-1e9c-48be-8944-0bce2c6f441d/6-1940

In `IntegralAP::truncate()` (which we call when getting the `3` and setting it as the value of the bitfield), we actually truncate the bitwith of the underlying `IntAP`, not just the _value_.

This fixes things for me:
```diff
diff --git a/clang/lib/AST/Interp/IntegralAP.h b/clang/lib/AST/Interp/IntegralAP.h
index bab9774288bf..fb7ee1451571 100644
--- a/clang/lib/AST/Interp/IntegralAP.h
+++ b/clang/lib/AST/Interp/IntegralAP.h
@@ -154,7 +154,10 @@ public:
   }

   IntegralAP truncate(unsigned BitWidth) const {
-    return IntegralAP(V.trunc(BitWidth));
+    if constexpr (Signed)
+      return IntegralAP(V.trunc(BitWidth).sextOrTrunc(this->bitWidth()));
+    else
+      return IntegralAP(V.trunc(BitWidth).zextOrTrunc(this->bitWidth()));
   }

   IntegralAP<false> toUnsigned() const {
```
I'd push this (it doesn't break any existing tests), but I'm not 100% sure if this is the right way to do that. Also, I should follow up with a `truncate` -> `truncateValue` rename.

https://github.com/llvm/llvm-project/pull/89254


More information about the cfe-commits mailing list