Re: the mangling of ‘va_list’ has changed in GCC 4.4

[Riku Voipio <riku.voipio@iki.fi>]

Modestas Vainius wrote:
> I have recently came across .symver directive [1][2] that is more 
> straightforward way to achieve the same thing. In that case, only source file 
> (.cpp) needs to be changed. 
Brilliant! This shall be the plan then:

1) identify affected libraries (Jakubs list as base?)
2) provide patches for the libraries and submit them to BTS
3) give maintainers X days to MU the libraries, after that proceed to 
NMU them
4) binNMU any packages using the functions with va_list parameters
5) release squeeze ;)
6) tell maintainers they are free to drop the symbol alias

Sounds ok to everyone?

> For example:
>
> -----------------------------------------------
> $ cat test.h
> void myprint(const char* fmt, va_list list);
>
> $ cat libtest.cpp
> #include <cstdarg>
> #include "test.h"
>
> void myprint(const char* fmt, va_list list) {
> }
>
> // va_list mangling has been changed in g++ 4.4
> // Add ABI <= g++-4.3 compatibility hack
> #if defined __ARM_EABI__ && defined __GNUC__ && (__GNUC__ > 4 || __GNUC__ == 4 
> && __GNUC_MINOR__ >= 4)
> __asm__(".symver _Z7myprintPKcSt9__va_list, _Z7myprintPKcPv@@");
> #endif
> -----------------------------------------------
>
> "_Z7myprintPKcSt9__va_list" ( myprint(char const*, std::__va_list) )
>  here is the new mangling while "_Z7myprintPKcPv ( myprint(char const*, void*) 
> )" is the old mangling. When this code is built under g++-4.4, the resulting 
> binary will have both symbols with _Z7myprintPKcPv and 
> '_Z7myprintPKcSt9__va_list' referring to the same address:
>
> $ objdump -T libtest.so | grep _Z7myprint
> 000006d8 g    DF .text  00000058  Base        _Z7myprintPKcSt9__va_list
> 000006d8 g    DF .text  00000058  Base        _Z7myprintPKcPv
>
> 1. http://www.redhat.com/docs/manuals/enterprise/RHEL-3-Manual/gnu-
> assembler/symver.html
> 2. http://www.redhat.com/docs/manuals/enterprise/RHEL-4-Manual/gnu-
> linker/version.html
>
>