PR27128, nm -P portable output format regression

[Fangrui Song]

On 2020-12-31, Alan Modra via Binutils wrote:
>On Wed, Dec 30, 2020 at 06:10:34AM -0800, H.J. Lu wrote:
>> I think they are useful.   Here is a patch to add tests.
>
>Fails on mips due to mips overriding .byte and then foo and _Zrm1XS_
>having the same address.  Which is quite weird.  This results in -n
>sort not ordering the symbols as expected.  I'm modifying the tests to
>use .space 16 instead.
>
>Also, this
>> +VERS_2.0 +|0+| +A +| +|0+1|.*
>matches anything.  I'm adding backslash escapes to match literal '|'
>and '.', plus correcting the sysv expected lines.
>
>Committed.
>
>binutils/
>	PR 27128
>	* nm.c (print_symname): Append version string to symbol name
>	before printing the lot under control of "form".  Append version
>	to demangled names too.
>ld/
>	PR 27128
>	* testsuite/ld-elf/pr27128.s: New file.
>	* testsuite/ld-elf/pr27128.t: Likewise.
>	* testsuite/ld-elf/pr27128a.d: Likewise.
>	* testsuite/ld-elf/pr27128b.d: Likewise.
>	* testsuite/ld-elf/pr27128c.d: Likewise.
>	* testsuite/ld-elf/pr27128d.d: Likewise.
>	* testsuite/ld-elf/pr27128e.d: Likewise.

Oh, I did not know "gas: Extend .symver directive" (`.symver ... , remove`) has been committed.

`remove` will pretty much be always preferred
(https://maskray.me/blog/2020-11-26-all-about-symbol-versioning#Assembler-behavior)

However, it can be confusing. Instead of adding an optional argument to .symver, what about adding a new directive?



Note that copying attributes (.symver) isn't traditional behavior: for .set foo, bar, the visibility/binding bits of foo and bar are separate.
The new directive can just do renaming.