On Monday 19 August 2024 13:50:14 Martin Storsjö wrote: > On Sat, 17 Aug 2024, Pali Rohár wrote: > > > On Monday 05 August 2024 17:00:20 LIU Hao wrote: > > > 在 2024-08-05 00:44, Pali Rohár 写道: > > > > _init_scprintf: > > > > ... > > > > .def __vscprintf; .scl 2; .type 32; .endef > > > > .def _init_scprintf; .scl 2; .type 32; .endef > > > > > > > > It is correct that _init_scprintf has two ".def" lines? > > > > > > Probably not... If it is declared elsewhere, we might want to remove the > > > `.def` directive from assembly. > > > > > > And whatever scl (storage class) it appears in the symbol table probably > > > does not matter. The symbol shall not escape that .o file, and that's all. > > > > > > > > > -- > > > Best regards, > > > LIU Hao > > > > I run 'nm' on the original version and new version (without static). The > > original version is: > > > > 00000000 b .bss > > 00000000 d .data > > 00000010 t _emu_scprintf > > U __imp__GetModuleHandleW@4 > > U __imp__GetProcAddress@8 > > 00000000 D __imp___scprintf > > 00000000 t _init_scprintf > > 00000000 r .rdata > > 00000000 r .rdata$zzz > > 00000030 t _resolve_scprintf > > 00000007 T __scprintf > > 00000000 t .text > > U __vscprintf > > > > And the difference between original a new is: > > > > @@ -4,7 +4,7 @@ > > U __imp__GetModuleHandleW@4 > > U __imp__GetProcAddress@8 > > 00000000 D __imp___scprintf > > -00000000 t _init_scprintf > > +00000000 T _init_scprintf > > 00000000 r .rdata > > 00000000 r .rdata$zzz > > 00000030 t _resolve_scprintf > > > > When static is removed then init_scprintf is marked with uppercase T, > > same as _scprintf. So it looks like that for GNU binutils, the > > init_scprintf is global / exported when static keyword is not present. > > > > So it looks like that the symbol can escape from the .o file and can be > > referenced by other files. > > Indeed. > > But the original issue - a function which on the C level is declared static, > but actually isn't defined anywhere in the C code, sounds a bit problematic > indeed; I would not be surprised if the compiler would complain about it, > possibly in LTO modes as well. And if not now, then possibly in future > compiler versions.
Is not there a documented way how to write such code for gcc and clang? Because inline assembly in C code is supported for both gcc and clang. > I think the safest way to handle it would be to make the symbol non-static, > which would require giving it a prefix starting with __ (and maybe a full > __mingw prefix). > > // Martin _______________________________________________ Mingw-w64-public mailing list Mingw-w64-public@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
