Re: QNX binutils targets

[Hans-Peter Nilsson <hp at bitrange dot com>]

On Fri, 25 Oct 2002, H. J. Lu wrote:
> On Fri, Oct 25, 2002 at 01:13:07PM -0400, Hans-Peter Nilsson wrote:
> > Consider instead creating a special-named section, as proposed
> > in <URL:http://sources.redhat.com/ml/binutils/2002-10/msg00454.html>.
>
> You have no ideas what you are talking about.

Yes I do.  (Is this one of those flame wars? :-)

Please avoid sounding condescending, at least on the subject of
a debated issue.

I'm considering compatibility with existing tools and objects
too, which I think you're not with the ELFOSABI change.  Having
the assembler always emit a special section by which the linker
can resolve (otherwise) ambiguity is a good solution IMHO.

> > If you follow the binutils list, you should know that H.J.:s
> > proposal to change ELFOSABI for this purpose is doubtful.
> >
> > See for example
> > <URL:http://sources.redhat.com/ml/binutils/2002-10/msg00434.html>,
> > links and follow-ups in/to that message.
>
> You totally missed the boat. EI_OSABI is for linker

It is, but with respect to the *semantics* of other ELF fields,
so not applicable here.  Letting the linker check objects by
e.g. presence of a .IA32.abi.qnx section seems like a working
solution to me.  It's what Uli suggests, IIUC!

> and .note.ABI-tag
> is for OS.

Agreed, never debated.

> Please see
> http://sources.redhat.com/ml/binutils/2002-10/msg00436.html

I did (see below); it indicates that a EI_OSABI/ELFOSABI change
is wrong.

> BTW, Ulrich and I were therre when we, people who are responsible
> for gABI and psABI/ia64, aka ELF, discussed how EI_OSABI should be
> used.

> Before you mislead anyone again,

Right back atcha!

> please read
>
> http://sources.redhat.com/ml/binutils/2000-11/msg00383.html

I did, again!  (And I pointed to it, indirectly, too in the
mesage you replied to!) Perhaps you should read it too, again!

Anyway, it still indicates that it EI_OSABI the *interpretation
of other fields* (that is, what SHT_LOOS through SHT_HIOS means)
not *what system the object is for*.  Despite the name of the
field: ABI refers to the meaning of ELF fields, not of the
binary contents.  It can technically be abused for that purpose,
(a fast and simple solution, I agree) but besides being
spec-wise wrong, it breaks with previous tools and objects,
which is bad, I think.

Please correct me if I'm wrong.
(Specifically directed as an invitation to other people -- too.)

brgds, H-P