Issues with endianess: test pixel_iterator fails

[trex58]

Actual behavior

v1 = 5

This deals with what were described in 2010 at: https://lists.boost.org/Archives/boost/2010/09/170446.php

On AIX (Big Endian), test pixel_iterator (and probably others) fail.

I have:

    Assertion failed: v1 == 6, file ../libs/gil/test/pixel_iterator.cpp, line 137

Instead of v1=6, I have v1=5. Values for vI, I=2..8, are wrong too.

About p1/v1, this appears to be due to:

    ../boost/gil/channel.hpp:368 : return ret;

ret=46701 = 0xb66d : d/16bit = 13/10 = 1101/2 : 3 last bits: 101/2 = 5 !!!

The expected correct value (as seen on Fedora/x86_64) is:

ret=28086 = 0x6db6 : 6/16bit = 6/10 = 0110/2 : 3 last bits: 110/2 = 6 !!!

On AIX, bytes are inverted for "ret" : b6 6d instead of: 6d b6.

I'm looking for the parts of the code to be made endianess-aware ??

Possibilities:

  ? gil/bit_aligned_pixel_reference.hpp : bit_range& operator++() { ... } ?
  ? gil/channel.hpp : struct static_copy_bytes { ... } ?
  ? gil/channel.hpp : integer_t get() const { ... } ?

Value of p1:

AIX: (gdb) p p1
$68 = {static bit_size = 3, static is_mutable = <optimized out>, _bit_range = {_current_byte = 0x110022510 "\266m\333", _bit_offset = 0}}
x 0x110022510 :: 0xb66ddb00

Linux: (gdb) p p1
$21 = {static bit_size = 3, static is_mutable = <optimized out>, _bit_range = {_current_byte = 0x41beb0 "\266", <incomplete sequence \333>, _bit_offset = 0}}
(gdb) x 0x41beb0 :: 0x00db6db6

Expected behavior

v1 = 6

C++ Minimal Working Example

test pixel_iterator

Environment

• Operating System: AIX 7.2
• Compiler version: GCC 8.3.0
• Build settings: 64bit
• Version (Git ref or <boost/version.hpp>): 1.69.0