ALIGN directive showing different behavior than documented

[Andre Vieira]

Hello,

Andreas Fritiofson reported at 
https://bugs.launchpad.net/gcc-arm-embedded/+bug/1471226 an issue 
regarding the ALIGN directive in linker scripts. As he points out the 
documentation says "ALIGN(align) is equivalent to ALIGN(., align)".

See below an expanded version of his example.

echo > test.c

cat test.ld
align = 0x100;

SECTIONS
{
    .section_with_unaligned_end :
      {
      test_section1 = .;
      . += 0x1234;
      }
    .test :
      {
      test0 = .;
      . += 0xabcd;
      test1 = .;
      . = ALIGN(align);
      test2 = .;
      test3 = ALIGN(., align);
      . = ALIGN(align);
      test4 = .;
      . += 0x22;
      test6 = ALIGN(., align);
      test7 = ALIGN(align);
      test8 = ALIGN(., align);
      }
    /DISCARD/ : { *(*) }
}

$ arm-none-eabi-gcc test.c -T test.ld

$ arm-none-eabi-nm a.out | grep test
00001234 B test0
0000be01 B test1
0000bf00 B test2
0000bf34 B test3
0000bf00 B test4
0000bf34 B test5
00000000 B test_section1

As you can see I expanded his test with a couple more ALIGN calls to do 
a bit more exploring. They are obviously different so the documentation 
is misleading at best. Furthermore, as Andreas points out it seems that 
the binary operator calculates it's alignment relative to the start of 
the section. In this example, the start of section test this is 0x1234.

After some digging I found that the unary operator uses the expression 
'expld.dot' whereas the binary operator uses 'lhs.value'. The first 
gives an absolute value, whereas the second seems to hold a relative 
value to the section start.

The fact that the unary operator uses new_rel_from_abs did raise some 
flags with me. Not entirely sure what the function is doing there.

This leaves with some uncertainties to whether it's just a documentation 
issue or a faulty code. Please advise!

Thank you in advance for the time spent on this, I know your plates are 
already quite full!

Kind Regards,
Andre Vieira