Re: [PATCH] Fix distinction of 32/64bit addresses in MIPS gas

["H . J . Lu" <hjl@lucon.org>]

On Thu, Sep 06, 2001 at 11:38:31AM -0700, cgd@broadcom.com wrote:

> Yes, something might be a bug w.r.t. an ABI standard that nobody
> started following until 6 months ago.  But e.g. some of the changes to

That is why I said the mips ELF implementation in binutils was only
really tested on Linux/mips :-).

> binutils will cause binary compatibility for NetBSD/mips binaries,
> which have been using the old MIPS ELF ABI implemented by binutils for
> ... 3+ years.
> 
> I don't disagree that bugs need to be fixed, but some of the fixes
> will cause some groups real pain.
> 

Been there, done that. It happened quite a few times on some Linux
platforms. Sorry, I can't help on that.

> > The only mips testsuite results for gcc I can find are Irix and Linux.
> > It won't surprise me that the next release of gcc/binutils won't work
> > on NetBSD/mips, out of box.
> 
> GCC isn't relevant here, is it?  I mean, binutils is supposed to be a
> usable compiler with gcc, without gcc, and also with random
> (not-current) versions of gcc.  NetBSD is ... much less likely to
> upgrade to a new version of gcc than to a new version of binutils.

It depends on how you see it. In my view, a toolchain should at least
include binutils and gcc. They should be compatible with each other.

> The point of my message was mostly informational: i think the list
> should be aware that there are issues re: MIPS ABIs, current binutils,
> and compatibility with a system that's been using binutils to generate
> dynamically-linked MIPS ELF binaries for years.
> 

Well, it is not an excuse not to have a good implementation of the SVR4
MIPS ABI.

BTW, "make check" in binutils checks the SVR4 ELF ABI implementation on
Linux/mips. Linux/mips also has its own testsuite for the SVR4 ELF ABI
implementation in glibc. Does NetBSD/mips have similar tests?


H.J.