[PATCH] Improve ptrace-error detection on Linux targets

[Pedro Alves]

Two comments on implementation, that I think apply 
equally to the new version you're working on as well:

#1 - A target method (target_ptrace_me_fail_reason) looks like the
  wrong place to put this, to me.  As I understand it, the core of
  gdb isn't ever calling the target_ptrace_me_fail_reason.  This is
  all code that is implementation detail of the Linux target
  implementation.

  I'd prefer some other way to address this.  An idea would be to have
  a function pointer in inf-ptrace.c that defaults to an implementation
  that returns the empty string:

   // in inf-ptrace.c

   static std::string
   default_inf_ptrace_me_fail_reason (int err)
   {
     return {};
   }

   std::string (*inf_ptrace_me_fail_reason) (int err) = default_inf_ptrace_me_fail_reason;

   // in inf-ptrace.h

   extern std::string (*inf_ptrace_me_fail_reason) (int err);

   And then in linux-nat.c:_initialize_linux_nat, you'd override the
   implementation, like:

   std::string
   linux_ptrace_me_fail_reason (int err)
   {
     ...
   }

   void
   _initialize_linux_nat ()
   {
      ... 
      inf_ptrace_me_fail_reason = linux_ptrace_me_fail_reason;
   }


  Note I'm not suggesting anything with virtual methods before
  of the next point.

#2 - What about gdbserver?  Don't we want the same nice error
messages in gdbserver?

See gdbserver/linux-low.c:linux_ptrace_fun.

This would suggest putting that linux_ptrace_me_fail_reason
function in gdb/nat/ so that it would be used by gdbserver too.

Thanks,
Pedro Alves