https://github.com/DaanishD20 created
https://github.com/llvm/llvm-project/pull/207730
Fixes #200139.
## Summary
An array new-expression whose bound is neither specified nor deducible
from its initializer (for example `new int[]()`) is ill-formed and is
already diagnosed by Sema with `err: cannot determine allocated array
size from initializer`. However, the constant evaluator still tried to
evaluate the recovery expression. In `PointerExprEvaluator::VisitCXXNewExpr`
the array size is unavailable (`E->getArraySize()` is `nullopt`), so the
allocation fell through to the scalar-new path, producing an array
`APValue` stored under a scalar type. Reading that object later asserted
in `Type::castAsArrayTypeUnsafe()` (`isa<ArrayType>(CanonicalType)`) from
`CheckEvaluationResult`.
This patch detects an array new-expression whose size could not be
determined and bails out of constant evaluation, so the already-emitted
diagnostic stands and no crash occurs. Well-formed array new-expressions
are unaffected: an explicit bound (`new int[3]()`) or a bound deduced from
a braced initializer (`new int[]{1, 2, 3}`) both keep a non-null array
size and continue to evaluate as before.
## Reproducer (reduced by @efriedma-quic in the issue)
```c++
static_assert((new int[]())[0] == 0);
```
Before this change, an assertions build of clang crashed in
`Type::castAsArrayTypeUnsafe()`. After it, clang reports the error and
exits cleanly.
## Testing
- Added a regression test to `clang/test/SemaCXX/constant-expression-cxx2a.cpp`.
- Confirmed the reduced and the original fuzzer reproducers from the issue
no longer crash.
- `llvm-lit clang/test/SemaCXX/constant-expression-cxx2a.cpp
clang/test/SemaCXX/new-delete.cpp clang/test/AST/ByteCode/new-delete.cpp` all
pass.
>From c7c76f1ed849f707f16f16a56260d79a35fc47e0 Mon Sep 17 00:00:00 2001
From: DaanishD20 <[email protected]>
Date: Mon, 6 Jul 2026 18:37:14 +0530
Subject: [PATCH] [clang] Fix constant-evaluator crash on array new with no
size
An array new-expression whose bound is neither specified nor deducible
from its initializer (e.g. `new int[]()`) is ill-formed and is diagnosed
by Sema, but the constant evaluator still tried to evaluate the recovery
expression. It treated the array allocation as a scalar one, producing an
array APValue stored under a scalar type, and eventually asserted in
`Type::castAsArrayTypeUnsafe()` while checking the evaluation result.
Detect an array new-expression whose size could not be determined in
`PointerExprEvaluator::VisitCXXNewExpr` and bail out gracefully instead
of crashing.
Fixes #200139.
---
clang/docs/ReleaseNotes.md | 1 +
clang/lib/AST/ExprConstant.cpp | 7 +++++++
clang/test/SemaCXX/constant-expression-cxx2a.cpp | 6 ++++++
3 files changed, 14 insertions(+)
diff --git a/clang/docs/ReleaseNotes.md b/clang/docs/ReleaseNotes.md
index b7142ecb072ff..e38b7c3b02caf 100644
--- a/clang/docs/ReleaseNotes.md
+++ b/clang/docs/ReleaseNotes.md
@@ -750,6 +750,7 @@ latest release, please see the [Clang Web
Site](https://clang.llvm.org) or the
- Fixed an assertion where we improperly handled implicit conversions to
integral types from an atomic-type with a conversion function. (#GH201770)
- Fixed assertion failures involving code completion with delayed default
arguments and exception specifications. (#GH200879)
- Fixed a regression where calling a function that takes a class-type
parameter by value inside `decltype` of a concept could be incorrectly rejected
when used as a non-type template argument. (#GH175831)
+- Fixed a crash in the constant evaluator when an ill-formed array
new-expression whose bound could not be determined (e.g. `new int[]()`) was
used in a constant expression. (#GH200139)
#### Bug Fixes to Compiler Builtins
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index fb82e0addfdac..bd188cd10f82f 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -10863,6 +10863,13 @@ bool PointerExprEvaluator::VisitCXXNewExpr(const
CXXNewExpr *E) {
AllocType = Info.Ctx.getConstantArrayType(AllocType, ArrayBound, nullptr,
ArraySizeModifier::Normal, 0);
+ } else if (E->isArray()) {
+ // We have an array new-expression whose array size could not be
+ // determined, e.g. 'new int[]()', where the bound is neither given nor
+ // deducible from the initializer. This is ill-formed and already
+ // diagnosed, so bail out rather than mis-evaluating a scalar allocation
+ // as an array (which would later crash the evaluator).
+ return false;
} else {
assert(!AllocType->isArrayType() &&
"array allocation with non-array new");
diff --git a/clang/test/SemaCXX/constant-expression-cxx2a.cpp
b/clang/test/SemaCXX/constant-expression-cxx2a.cpp
index 396a8df21a3e3..94ecafc081302 100644
--- a/clang/test/SemaCXX/constant-expression-cxx2a.cpp
+++ b/clang/test/SemaCXX/constant-expression-cxx2a.cpp
@@ -898,6 +898,12 @@ namespace dynamic_alloc {
static_assert(erroneous_array_bound(-1)); // expected-error {{constant
expression}} expected-note {{in call}}
static_assert(erroneous_array_bound(1LL << 62)); // expected-error
{{constant expression}} expected-note {{in call}}
+ // An array new-expression whose bound is neither given nor deducible from
the
+ // initializer is ill-formed; the constant evaluator must reject it
gracefully
+ // rather than crash. See GH200139.
+ static_assert((new int[]())[0] == 0); // expected-error {{cannot determine
allocated array size from initializer}} \
+ // expected-error {{static assertion
expression is not an integral constant expression}}
+
constexpr bool erroneous_array_bound_nothrow(long long n) {
int *p = new (std::nothrow) int[n];
bool result = p != 0;
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
