FreeBSD endian.hpp

[nrv2]

Hi, Thanks for providing such a great library. I have no complaints, but I have one little glitch that showed up for me... When building liblas on two FreeBSD systems, one i386, and one amd64, I ended up with a library that reads/writes in big-endian. After some investigation, the file include/liblas/detail/endian.hpp has the following lines appear after the initial test to see if GLIBC is defined (which it is not for me):

...
#elif defined(_BIG_ENDIAN)
# define LIBLAS_BIG_ENDIAN
# define LIBLAS_BYTE_ORDER 4321
#elif defined(_LITTLE_ENDIAN)
# define LIBLAS_LITTLE_ENDIAN
# define LIBLAS_BYTE_ORDER 1234
... 

However, on my system both _BIG_ENDIAN and _LITTLE_ENDIAN are defined, as 4321 and 1234 respectively. So the first test for _BIG_ENDIAN wins, and this results in LIBLAS_BIG_ENDIAN being defined and a big-endian reader/writer. All is well if I comment out these six lines. In the header, the following tests for architecture definitions work correctly for both my systems. I did not get to try it on any other systems/OSes.

This was a pretty simple fix, and I already sent patches to the FreeBSD port maintainer for liblas. So if that code is necessary for other OS types, no change is needed. However, it would be great if no changes were needed to build on FreeBSD in the future.

I'm attaching my slightly modified endian.hpp in case the description above is not clear.

Thanks, Nick

[nrv2]

Modified endian.hpp to allow proper build on FreeBSD

[hobu]

Thanks for the report. This one is Mateusz' :)

[mloskot]

Nick's fix in form of patch generated against r1246

[mloskot]

Nick,

Thanks for the patch, however I can not accept it because I'm not sure what it fixes actually. It just comments out important checks, is that right?

AFAIU, on FreeBSD (on your machine) there is 3rd situation possible which is currently not handled by libLAS, it is both _BIG_ENDIAN and _LITTLE_ENDIAN macros are defined. Am I correct?

If it is correct, then I'd suggest to fix according to this scheme - 3 cases:

#if defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN) // FreeBSD + Nick's machine
... // What to set? Nothing?
#else
# if defined(_BIG_ENDIAN)      // Big endian only
#  define LIBLAS_BIG_ENDIAN
#  define LIBLAS_BYTE_ORDER 4321
# elif defined(_LITTLE_ENDIAN) // Little endian only
#  define LIBLAS_LITTLE_ENDIAN
#  define LIBLAS_BYTE_ORDER 1234
# endif
#endif

[nrv2]

Sorry for the time delay, I did not realize there was another post. You understand the situation correctly. Both _BIG_ENDIAN and _LITTLE_ENDIAN macros are defined. Not sure what happens on a big endian system running FreeBSD, but your suggestion in the last post would work fine on my machines.

[hobu]

I have attempted to fix this by testing for the case where both _BIG_ENDIAN and _LITTLE_ENDIAN are defined and just assuming LITTLE.

r1323 and r1324