================
@@ -564,6 +564,9 @@ unsigned Parser::ParseAttributeArgsCommon(
nullptr,
Sema::ExpressionEvaluationContextRecord::EK_AttrArgument);
+ SaveAndRestore<ConversionAction> SavedTranslationState(
+ ParserConversionAction, CA_NoConversion);
----------------
s-barannikov wrote:
What follows is just my thoughts, which may not align well with the standard.
I consider an *evaluated* string to be a string that can be used by the
compiled program at runtime.
-fexec-charset controls the encoding of such strings.
`asm` statements and the `#pragma map` (I suppose it is vendor-specific) are
clearly not supposed to be used by the program. Instead, they are processed by
the host tools (assembler/linker). I agree `-fexec-charset` should not be
applicable to these string literals, which (in my mindset) means they should be
*unevaluated*.
Unevaluated strings do not have encoding because they have no "value". Note
that the standard specifically forbids numerical escape sequences that would
give value to a character. They are not supposed to make their way into a
compiler artifact (be it an asm file or an object file). In practice, they do.
Now, if we want to treat an asm string as evaluated, it should respect
`-fexec-charset` option. Otherwise, it cannot contain octal escapes. If neither
of that is acceptable, I guess we need to make the desired behavior
customizable.
`#pragma map` should have a handler where we can do whatever we want. I don't
see a problem here, but I may miss something.
Regarding asm statements, I would very much prefer them be unevaluated. I'm not
sure I understand why do you use escape sequences in asm strings. Does the
(external) assembler expect EBCDIC input? If so, why only strings from asm
statements are EBCDIC-encoded and not the rest of the compiler-generated
assembly file?
https://github.com/llvm/llvm-project/pull/169803
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
