[Clang] disallow non-lvalue values in constant expressions to prevent invalid pointer offset computation

[a-tarasyuk]

Fixes #95366

[llvmbot]

@llvm/pr-subscribers-clang

Author: Oleksandr T. (a-tarasyuk)

Changes

Fixes #95366

---

Full diff: https://github.com/llvm/llvm-project/pull/95479.diff

3 Files Affected:

• (modified) clang/docs/ReleaseNotes.rst (+1)
• (modified) clang/lib/AST/ExprConstant.cpp (+3)
• (added) clang/test/Sema/integral-to-ptr.c (+3)

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index 8c2f737836a9d..755557906360b 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -847,6 +847,7 @@ Bug Fixes to C++ Support
 - Fixed several bugs in capturing variables within unevaluated contexts. (#GH63845), (#GH67260), (#GH69307),
   (#GH88081), (#GH89496), (#GH90669) and (#GH91633).
 - Fixed handling of brace ellison when building deduction guides. (#GH64625), (#GH83368).
+- Fix an assertion failure caused by non-lvalue usage in lvalue context. (GH95366).
 
 Bug Fixes to AST Handling
 ^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 7178f081d9cf3..08bee806f172f 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -9325,6 +9325,9 @@ bool PointerExprEvaluator::VisitCastExpr(const CastExpr *E) {
       Result.IsNullPtr = false;
       return true;
     } else {
+      if (!Value.isLValue())
+        return false;
+
       // Cast is of an lvalue, no need to change value.
       Result.setFrom(Info.Ctx, Value);
       return true;
diff --git a/clang/test/Sema/integral-to-ptr.c b/clang/test/Sema/integral-to-ptr.c
new file mode 100644
index 0000000000000..99f83c3e52057
--- /dev/null
+++ b/clang/test/Sema/integral-to-ptr.c
@@ -0,0 +1,3 @@
+// RUN: %clang_cc1 %s -verify -fsyntax-only -std=c11
+
+int x(void) { e: b: ; return &&e - &&b < x; } // expected-warning {{ordered comparison between pointer and integer ('long' and 'int (*)(void)')}}

[Sirraide]

LGTM after addressing one issue.

Looking at the rest of the constant evaluator (and the comments on the issue as well), everything else seems to use EvaluateInteger, which makes sure the result of the evaluation is an integer, so this seems like it is the only place where we can wind up with an AddrLabelDiff, so rejecting it here makes sense.

[cor3ntin]

Can you make sure the PR/commit have titles and messages that follow our guidelines? Thanks!

[AaronBallman]

Changes generally LGTM though I had some suggestions to consider.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[Sirraide]

There’s a typo in the release notes, but everything else lg.

[AaronBallman]

LGTM!