Page History

...

• The PX4 team uses these patches to cwhar and math.h to solve the issue. But note the comments in that code:

    /* N.B. The following definitions are enabled at this time to allow the PX4 
 * development to continue until there is a SAFE  solution to foreign

...

 
 * (non-nuttx) header file inclusion. There is a potential of a binary

...

 
 * incompatibility and runtime errors, memory overwrites or corruption

...

 
 * VVVVVVVVVVVVVVVVVVVVVVVVVV Begin Warning VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV

...

 
 */ 

• Some people have suggested adding the type definition of wint_t to nuttx/include/sys/types.h merely because that header file will then be included into the newlib math.h. This inclusion, of course, also very dangerous since the types in the NuttX sys/types.h header file may not agree with the types in the pre-compiled newlib math library. This solution is not recommended, in any case. The type wint_t is already correctly defined in nuttx/include/sys/wchar.h which is the one and only correct location per OpenGroup.org. It is a mystery to my why the newlib math.h header file uses wint_t without including wchar.h. If it did, then there would then this compilation issue would not exist (there could still be subtle binary compatibility issues).

...

  commit 894ca622e6a408e5fa858a3fee46fb16f32cf86c
  Author: Xiang Xiao <xiaoxiang@xiaomi.com>
  Date:   Mon Aug 27 06:26:37 2018 -0600

    include/sys/types:  Move wint_t and wctype_t from wchar.h to
    types.h.  This change is compatible as before since wchar.h
    include types.h indirectly.  This fixes a compilation error with
    newlib's math.h:  'unknown type name wint_t'

...

  #ifndef _ALLOCA_H
  #define _ALLOCA_H

  #define alloca __builtin_alloca

  #endif /* _ALLOCA_H */

And add the path to this alloca.h to your CFLAG include path arguments. The path can specified by adding -system or -I to the CFLAGS. The path to this alloca.h must be defined last so that it has precedence.

...

  #ifndef _MYMATH_H
  #define _MYMATH_H
  #ifndef _ALLOCA_H
  #define _ALLOCA_H

  #define alloca __builtin_alloca

  #endif /* _ALLOCA_H */

  #include_next <math.h>

  #endif /* _MYMATH_H */

This will provide the alloca() definition, then continue to include the default version of math.h. This works because the idempotence pre-processor variable _ALLOC_H matches the same idempotence variable used in the newlib alloca.h. Thus, any sneak inclusion of alloca.h with have not effect.

...