Re: Question on enabling c++ to_string to arm-none-eabi target

[Jonathan Wakely <jwakely dot gcc at gmail dot com>]

On 31 October 2013 10:10, Terry Guo wrote:
>
> The stdio.h from Newlib defines vscanf function as:
>
> #if !defined(__STRICT_ANSI__) || (__STDC_VERSION__ >= 199901L)
> int     _EXFUN(vscanf, (const char *, __VALIST)
>                _ATTRIBUTE ((__format__ (__scanf__, 1, 0))));
> #endif
>
> And the libstdc++-v3 configuration always uses option -std=c++98 to
> compile above code, this option indicate the define of
> __STRICT_ANSI__. So always fail to compile above code with error
> message:
>
> conftest.cpp:36:19: error: 'vscanf' was not declared in this scope
>   vscanf("%i", args);


We have to use -std=c++xx not -std=gnu++xx (and so define
__STRICT_ANSI__) because users might want to use that, and we have to
know if the functions are available in all cases.


> Finally failing to compile the conftest.cpp causes the disable of c++
> to_string function. So what should I do to let those conftest.cpp
> pass? By changing option from -std=c++98 to -std=gnu++98 or removing
> the IF macros from stdio.h? Please help. Thanks in advance.


Using -std=gnu++98 would not be correct, because the functions would
still not be defined if users compile with -std=c++11 (which also
defines __STRICT_ANSI__), so GCC would build OK but users would get
errors when they compile.

It has been suggested in the past that we might want to split up the
_GLIBCXX_USE_C99 test to be more fine-grained.  For example, C99
features needed by C++11-only features (such as std::to_string) could
test using -std=c++11, which sets __cplusplus to 201103L, and a
conforming C library should enable the full C99 library in that case
(it looks like newlib wouldn't though, that's a bug in newlib.)

Alternatively, we could predefine __STDC_VERSION__ to 199901L to
coerce newlib into providing the C99 library (it's
implementation-defined whether that is defined by C++ implementations
or not.)