This is the mail archive of the
mailing list for the GDB project.
Re: [RFC] GDB ARIndex Linux rule cleanup
- From: "Ulrich Weigand" <uweigand at de dot ibm dot com>
- To: pedro at codesourcery dot com (Pedro Alves)
- Cc: gdb-patches at sourceware dot org, muller at ics dot u-strasbg dot fr (Pierre Muller), brobecker at adacore dot com ('Joel Brobecker'), gdb at sourceware dot org, eliz at gnu dot org ('Eli Zaretskii')
- Date: Wed, 15 Apr 2009 17:32:16 +0200 (CEST)
- Subject: Re: [RFC] GDB ARIndex Linux rule cleanup
Pedro Alves wrote:
> On Wednesday 15 April 2009 15:00:39, Ulrich Weigand wrote:
> > It seems to me that in *those* cases, we should actually use
> > GNU/Linux instead -- this is talking about GDB target support
> > for the GNU/Linux operating system on PowerPC, and not specifically
> > related to the kernel only (for example, those target descriptions
> > are also used when analysing a core file on a remote system).
> I disagree. Everything in that file is related to register descriptions
> and layouts defined by Linux, the kernel, what's in userland doesn't
> affect any of it, or did I miss something? Even if I built a system
> comprising of a Linux kernel + all BSD userland, these constants and
> offsets would still apply.
> > /* Create a prototype generic GNU/Linux target.=20=20
> I'm of the oposite opinion. I consider *this* a bug. This
> file (linux-nat.c) concerns about abstracting of ptrace and /proc
> interfaces, which are kernel defined interfaces.
See my reply to Eli ... lots of target-specific information is *not*
related to the Linux kernel, but the platform ABI, libc, libpthread ...
> > The other instances in your patch, where a kernel version
> > is specifically named, clearly refer to the Linux kernel.
> > I'd be fine with adding "kernel" to the text here; but I
> > also agree with Eli's suggestion that it might be even=20
> > better to have the ARI script recognize use of a version
> > number ...
> I re-suggest what Mark suggested, that we drop the ARI rule.
> It is only inventing work. I suggest we apply the Linux vs
> GNU/Linux judgement at patch review time, and be mostly careful
> in documentation and user visible strings, not in code directly
> interfacing with kernel data structures and interfaces.
I agree that user-visible instances are certainly much more
important, and I don't really have any strong opinion on whether
there is any need to enforce the GNU/Linux rule in source code
My point was simply that (to me, at least) there is clear evidence
of a pre-existing convention in GDB sources (all references to a
GDB "target" refer in fact to the target OS, and thus GNU/Linux
should be used), and those three places in ppc-linux* that Pierre
identified were the only deviations from that convention.
Dr. Ulrich Weigand
GNU Toolchain for Linux on System z and Cell BE