================
@@ -2231,6 +2231,45 @@ int foo() {
}
```
+### String literal encoding in ASM statements
+
+On targets where the platform's default ordinary literal encoding differs from
+UTF-8 (such as z/OS, where the system default encoding is IBM-1047):
+
+* **Direct ASM string literals**: Direct string literals used in `asm`
statements
+ or `asm` labels (such as `asm("...")` or `__asm__("...")`) are parsed and
+ interpreted using the target system's default encoding (such as IBM-1047 on
+ z/OS) rather than the literal encoding specified by `-fexec-charset`. This
+ ensures that escape sequences (such as octal or hexadecimal byte escapes)
+ are interpreted as system/assembler code points.
+
+* **Constexpr string expressions in ASM statements**: When a parenthesized
+ constant expression is provided in place of an `asm` string, string literals
+ within that constant expression are encoded according to the execution
+ character set specified by `-fexec-charset` (the literal encoding), as
+ constant expression evaluation expects `char` sequences to operate in the
+ literal encoding. The system encoding conversion applied to direct `asm`
+ string literals is restricted to the direct string literal case and does not
+ extend to string literals within constant expressions.
+
+For example, on z/OS compiled with `-fexec-charset UTF-8`:
+
+```c++
+constexpr std::string_view takeStringAndReturnEmptyStringView(const char *str)
{
+ return std::string_view();
+}
+
+int main(void) {
+ // Direct string literals in asm statements are interpreted in the system
encoding (IBM-1047).
+ asm("nop");
+
+ // String literals within the constexpr expression are encoded in the
execution charset (UTF-8).
+ // Characters not representable in IBM-1047 (such as "\N{LATIN CAPITAL
LETTER I WITH DOT ABOVE}")
+ // are valid here because constant evaluation evaluates the string literal
in UTF-8.
+ asm((takeStringAndReturnEmptyStringView("\N{LATIN CAPITAL LETTER I WITH DOT
ABOVE}")));
----------------
hubert-reinterpretcast wrote:
The wording is fine for what it covers; however, I think we should have some
agreement on how the result of the expression is interpreted before the PR
lands.
https://github.com/llvm/llvm-project/pull/169803
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
