[clang] Don't crash when -Wformat-signedness specified

[yicuixi]

Fixes #161075. This patch follow the changes in #150962, and fix a crash issue when -Wformat-signedness was specificed.

[llvmbot]

@llvm/pr-subscribers-clang

Author: None (yicuixi)

Changes

Fixes #161075

---

Full diff: https://github.com/llvm/llvm-project/pull/162049.diff

3 Files Affected:

• (modified) clang/docs/ReleaseNotes.rst (+2)
• (modified) clang/lib/Sema/SemaChecking.cpp (+3-1)
• (added) clang/test/Sema/format-strings-signedness.cpp (+30)

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index 46d56bb3f07f5..7952cbcad0dfc 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -366,6 +366,7 @@ Bug Fixes in This Version
 - Fixed an assertion when an improper use of the ``malloc`` attribute targeting
   a function without arguments caused us to try to access a non-existent argument.
   (#GH159080)
+- Fixed clang crash that caused by missing diagnostic argument. (#GH161072)
 
 Bug Fixes to Compiler Builtins
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -417,6 +418,7 @@ Bug Fixes to C++ Support
   ``__builtin_addressof``, and related issues with builtin arguments. (#GH154034)
 - Fix an assertion failure when taking the address on a non-type template parameter argument of
   object type. (#GH151531)
+- Fix an assertion failure of format string signedness check when ``-Wformat-signedness`` (#GH161075).
 
 Bug Fixes to AST Handling
 ^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 00f40cfa910d2..91834d430c987 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -8634,8 +8634,10 @@ CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
       case ArgType::Match:
       case ArgType::MatchPromotion:
       case ArgType::NoMatchPromotionTypeConfusion:
-      case ArgType::NoMatchSignedness:
         llvm_unreachable("expected non-matching");
+      case ArgType::NoMatchSignedness:
+        Diag = diag::warn_format_conversion_argument_type_mismatch_signedness;
+        break;
       case ArgType::NoMatchPedantic:
         Diag = diag::warn_format_conversion_argument_type_mismatch_pedantic;
         break;
diff --git a/clang/test/Sema/format-strings-signedness.cpp b/clang/test/Sema/format-strings-signedness.cpp
new file mode 100644
index 0000000000000..66bf1de1ec09a
--- /dev/null
+++ b/clang/test/Sema/format-strings-signedness.cpp
@@ -0,0 +1,30 @@
+// RUN: %clang_cc1 -triple=x86_64-pc-linux-gnu -fsyntax-only -verify -Wformat -Wformat-signedness %s
+// RUN: %clang_cc1 -triple=x86_64-pc-win32 -fsyntax-only -verify -Wformat -Wformat-signedness %s
+
+// Verify that -Wformat-signedness alone (without -Wformat) trigger the
+// warnings. Note in gcc this will not trigger the signedness warnings as
+// -Wformat is default off in gcc.
+// RUN: %clang_cc1 -triple=x86_64-pc-linux-gnu -fsyntax-only -verify -Wformat-signedness %s
+// RUN: %clang_cc1 -triple=x86_64-pc-win32 -fsyntax-only -verify -Wformat-signedness %s
+
+// Verify that -Wformat-signedness warnings are not reported with only -Wformat
+// (gcc compat).
+// RUN: %clang_cc1 -triple=x86_64-pc-linux-gnu -fsyntax-only -Wformat -verify=okay %s
+
+// Verify that -Wformat-signedness with -Wno-format are not reported (gcc compat).
+// RUN: %clang_cc1 -triple=x86_64-pc-linux-gnu -fsyntax-only -Wformat-signedness -Wno-format -verify=okay %s
+// RUN: %clang_cc1 -triple=x86_64-pc-linux-gnu -fsyntax-only -Wno-format -Wformat-signedness -verify=okay %s
+// okay-no-diagnostics
+
+// Ignore 'GCC requires a function with the 'format' attribute to be variadic'.
+#pragma GCC diagnostic ignored "-Wgcc-compat"
+namespace GH161075 {
+template <typename... Args>
+void format(const char *fmt, Args &&...args)
+    __attribute__((format(printf, 1, 2)));
+
+void do_format() {
+  bool b = false;
+  format("%hhi %hhu %hi %hu %i %u", b, b, b, b, b, b);  // expected-warning {{format specifies type 'unsigned char' but the argument has type 'bool', which differs in signedness}}
+}
+} // namespace GH161075

[yronglin]

LGTM

[erichkeane]

I thought that the report said this is trunk only? If this hasn't been released, teh release note isn't really meaningful.