https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85161
Jakub Jelinek <jakub at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |ASSIGNED Last reconfirmed| |2018-04-03 Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org Ever confirmed|0 |1 --- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Created attachment 43824 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43824&action=edit gcc8-pr85161.patch Untested fix. The problem is in incorrect macros being tested. The GCC (and clang) predefined macros for endianity are __BYTE_ORDER__, that needs to be compared against __ORDER_LITTLE_ENDIAN__ or __ORDER_BIG_ENDIAN__ or __ORDER_PDP_ENDIAN__. Then there are endian.h, sys/endian.h and/or machine/endian.h headers, which define depending on chosen feature test macros __BYTE_ORDER or BYTE_ORDER, which needs to be compared against __{BIG,LITTLE,PDP}_ENDIAN (or without the __ prefix depending on feature test macros). The code was comparing __BYTE_ORDER (normally defined by endian.h, sometimes included indirectly from stdlib.h depending on feature test macros) against __ORDER_BIG_ENDIAN (always undefined, so in preprocessor assumed to be 0). So, depending on whether endian.h was included or not, it was either never byte swapping (wrong on non-little endian) or byte swapping always (wrong on little endian).