Re: [PATCH 1/3] Add fbsd_nat_add_target.

[Mark Kettenis <mark.kettenis@xs4all.nl>]

> From: John Baldwin <jhb@freebsd.org>
> Date: Mon, 27 Apr 2015 15:16:50 -0400
> 
> On Monday, April 27, 2015 08:10:18 PM Pedro Alves wrote:
> > On 04/26/2015 02:24 AM, John Baldwin wrote:
> > > Add a wrapper for add_target in fbsd-nat.c to override target operations
> > > common to all native FreeBSD targets.
> > > 
> > > gdb/ChangeLog:
> > > 
> > > 	* fbsd-nat.c (fbsd_pid_to_exec_file): Mark static.
> > > 	(fbsd_find_memory_regions): Mark static.
> > > 	(fbsd_nat_add_target): New function.
> > > 	* fbsd-nat.h: Export fbsd_nat_add_target and remove prototypes for
> > > 	fbsd_pid_to_exec_file and fbsd_find_memory_regions.
> > > 	* amd64fbsd-nat.c (_initialize_amd64fbsd_nat): Use fbsd_nat_add_target.
> > > 	* i386fbsd-nat.c (_initialize_i386fbsd_nat): Likewise.
> > > 	* ppcfbsd-nat.c (_initialize_ppcfbsd_nat): Likewise.
> > > 	* sparc64fbsd-nat.c (_initialize_sparc64fbsd_nat): Likewise.
> > 
> > OOC, any reason you didn't instead do it like:
> > 
> > struct target_ops *
> > fbsd_nat_target (void)
> > {
> >   struct target_ops *t = inf_ptrace_target ();
> > 
> >   t->to_pid_to_exec_file = fbsd_pid_to_exec_file;
> >   t->to_find_memory_regions = fbsd_find_memory_regions;
> >   return t;
> > }
> > 
> > and then use fbsd_nat_target instead of inf_ptrace_target
> > directly?
> > 
> > This maps a little better to a C++ world.
> > 
> > linux-nat.c does it the way you did as it keeps a separate
> > linux_ops target instance around.
> 
> I was probably just using linux-nat.c as a reference.  One thing that
> confuses me about the linux-nat target is that it keeps linux_ops
> around so that it can call the original methods that it overrides,
> and yet for a few methods it also uses a local 'super_foo' variable
> to call an original method.  I think that those are both doing the
> same thing, but perhaps there is some subtlety I'm missing?
> 
> I do use a 'super_wait' to call ptrace's wait method in the second
> patch in this series, so I could certainly change this to return a
> target rather than modifying an existing one if that is preferred.

I'd say the linux-nat.c code is a bad example and recommend looking at
the obsd-nat.c code instead.  The linux-nat.c code is so complicated
because of all the workarounds needed to support threads.  The
linux_ops stuff is pretty much an artifact of those workarounds.  

I found that to add threads-support I did need to make modifications
to the _wait function that made it hard to re-use the
inf_ptrace_wait() code.  Sometimes code duplications just is the right
answer.