From d96569ecc2807a13dab6495d8cc4e82775b00af1 Mon Sep 17 00:00:00 2001
From: Pavel Skripkin <paskripkin@gmail.com>
Date: Fri, 2 Aug 2024 18:04:57 +0300
Subject: [PATCH] [analyzer] Fix crash on using `bitcast(<type>, <array>)` as
 array subscript (#101647)

Current CSA logic does not expect `LazyCompoundValKind` as array index.
This may happen if array is used as subscript to another, in case of
bitcast to integer type.

Catch such cases and return `UnknownVal`, since CSA cannot model
array -> int casts.

Closes #94496
---
 clang/lib/StaticAnalyzer/Core/Store.cpp | 12 +++++++++++-
 clang/test/Analysis/exercise-ps.c       | 20 ++++++++++++++++++--
 2 files changed, 29 insertions(+), 3 deletions(-)

diff --git a/clang/lib/StaticAnalyzer/Core/Store.cpp b/clang/lib/StaticAnalyzer/Core/Store.cpp
index 67ca61bb56ba28..b436dd746d21f0 100644
--- a/clang/lib/StaticAnalyzer/Core/Store.cpp
+++ b/clang/lib/StaticAnalyzer/Core/Store.cpp
@@ -472,7 +472,17 @@ SVal StoreManager::getLValueElement(QualType elementType, NonLoc Offset,
   const auto *ElemR = dyn_cast<ElementRegion>(BaseRegion);
 
   // Convert the offset to the appropriate size and signedness.
-  Offset = svalBuilder.convertToArrayIndex(Offset).castAs<NonLoc>();
+  auto Off = svalBuilder.convertToArrayIndex(Offset).getAs<NonLoc>();
+  if (!Off) {
+    // Handle cases when LazyCompoundVal is used for an array index.
+    // Such case is possible if code does:
+    //   char b[4];
+    //   a[__builtin_bitcast(int, b)];
+    // Return UnknownVal, since we cannot model it.
+    return UnknownVal();
+  }
+
+  Offset = Off.value();
 
   if (!ElemR) {
     // If the base region is not an ElementRegion, create one.
diff --git a/clang/test/Analysis/exercise-ps.c b/clang/test/Analysis/exercise-ps.c
index d1e1771afddb5e..50643d5b046879 100644
--- a/clang/test/Analysis/exercise-ps.c
+++ b/clang/test/Analysis/exercise-ps.c
@@ -1,10 +1,13 @@
-// RUN: %clang_analyze_cc1 %s -verify -Wno-error=implicit-function-declaration \
-// RUN:   -analyzer-checker=core,unix.Malloc \
+// RUN: %clang_analyze_cc1 %s -triple=x86_64-unknown-linux \
+// RUN:   -verify -Wno-error=implicit-function-declaration \
+// RUN:   -analyzer-checker=core,unix.Malloc,debug.ExprInspection \
 // RUN:   -analyzer-config core.CallAndMessage:ArgPointeeInitializedness=true
 //
 // Just exercise the analyzer on code that has at one point caused issues
 // (i.e., no assertions or crashes).
 
+void clang_analyzer_dump_int(int);
+
 static void f1(const char *x, char *y) {
   while (*x != 0) {
     *y++ = *x++;
@@ -30,3 +33,16 @@ void f3(void *dest) {
   void *src = __builtin_alloca(5);
   memcpy(dest, src, 1); // expected-warning{{2nd function call argument is a pointer to uninitialized value}}
 }
+
+// Reproduce crash from GH#94496. When array is used as subcript to another array, CSA cannot model it
+// and should just assume it's unknown and do not crash.
+void f4(char *array) {
+  char b[4] = {0};
+
+  _Static_assert(sizeof(int) == 4, "Wrong triple for the test");
+
+  clang_analyzer_dump_int(__builtin_bit_cast(int, b)); // expected-warning {{lazyCompoundVal}}
+  clang_analyzer_dump_int(array[__builtin_bit_cast(int, b)]); // expected-warning {{Unknown}}
+
+  array[__builtin_bit_cast(int, b)] = 0x10; // no crash
+}