Re: Support for alpha-osf, mips-irix

["Maciej W. Rozycki" <macro@linux-mips.org>]

On Thu, 9 Jul 2020, Rainer Orth wrote:

> Every once in a while I come across references to Tru64 UNIX
> (alpha*-dec-osf*) and SGI IRIX in the binutils-gdb tree.  This seems
> strange given that the last versions supported by GCC (Tru64 UNIX V5.1
> and IRIX 6.5) were obsoleted in GCC 4.7 back in 2012 and removed in GCC
> 4.8.  GDB took a bit longer, but GDB 7.9 removed support for both as
> well in 2015, although a few references have been overlooked.
> 
> Both platforms are long obsolete and I strongly suspect there's no one
> to benefit from continued support (whatever that means) in binutils alone.
> 
> Any objections to removing support completely?  I'm not yet certain how
> either or both is entangled with code for other targets still supported
> (alpha-linux, mips-linux).

 Technically we could delete the IRIX configuration triplets, but that 
does not buy us much as they use ELF (a peculiar variation of, e.g. the 
sort order of the symbol table is different from generic ELF), so painful 
extraction of dead pieces of ELF support code would be required, and then 
parts of IRIX 5 ELF format support, which the IRIX 6 ELF format has some
further differences from, are used by some embedded MIPS targets, so parts 
of that code would have to stay.  Not worth the effort IMO.

 Then IRIX test results look so-so mostly due to the ELF peculiarity 
causing output not to match regexps, but overall there's nothing wrong 
there AFAIK, just the missing manpower to clean the tests up (I did some 
of that work back when MIPS target maintenance was a part of my dayjob).  
Similar failures happen for the affected embedded targets, although 
they're limited to tests that do not require shared library support.

> There's even more prehistoric cruft, btw., e.g. references to
> mips-ultrix ;-)

 That (ECOFF) has been deleted AFAIK, except for minimal BFD support to 
keep `objcopy' to/from ELF working, to satisfy the requirement of some 
bootloaders fixed in the firmware.  Of course the removal of the remaining 
pieces made code verification a tad problematic (perhaps we could have a 
binutils/ test that runs `objcopy' both ways and checks basic consistency 
though).

  Maciej