[llvm-branch-commits] [clang] [clang-tools-extra] [llvm] Add format string handling (PR #196568)

Hubert Tong via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Sat Nov 14 21:03:44 PST 2026


================
@@ -17,11 +17,29 @@
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/FormatString.h"
 #include "clang/AST/Type.h"
+#include "clang/Basic/TargetInfo.h"
+#include "llvm/Support/TextEncoding.h"
 
 namespace clang {
 
 class LangOptions;
 
+/// Creates a TextEncodingConverter for format string character comparisons.
+/// On z/OS format strings are IBM-1047 encoded, so each byte must be converted
----------------
hubert-reinterpretcast wrote:

> The format string is stored in the literal encoding specified by -fexec-charset

A point in favour: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=20110 (still open) expresses the opinion that the format checking should be done based on the literal encoding.

> every target should convert characters back to the internal/host UTF-8 encoding before comparing them.

At least wholesale conversion of the format string from the literal encoding is not the correct mechanism. A conversion-based approach would fall apart because C's format strings are partially "character"-based, where a "character" means a byte. This is most noticeable in relation to the `[` conversion specifier of the `scanf` family, where a multibyte character whose encoding includes a byte matching the value of `]` can be used to specify the "characters" preceding the `]` as members of the scanset — some (but not all) C library implementations do behave this way. Also, conversion specifications in general are "character"-based, so redundant shift sequences that encode no characters should be observed within them (and not ignored) — this is actually also true when matching ordinary multibyte characters from the input (see https://www.open-std.org/jtc1/sc22/wg14/www/docs/dr_018.html).

https://github.com/llvm/llvm-project/pull/196568


More information about the llvm-branch-commits mailing list