Re: libio/gcc-3.0 status

[Roland McGrath]

> I tried to boot with -df (one of those, I don't remember which, makes sure
> init pauses before calling file_exec), and attach gdb to the task, and set
> break points.  

You have to make sure you're doing things in the proper order.  If you set
breakpoints before the file_exec happens, then gdb might just be in some
fantasy world thinking it wrote to memory that wasn't actually there.  You
need to attach, so that gdb bumps the task suspend count; then hit return
or ^J at init or whatever it is so the file_exec happens, then make gdb
notice ("info threads" or something) and then set the breakpoints.

> __task_terminate, which looks bogus.  The "backtrace" is bogus (I can
> repeat a typical gdb session here later).

info reg and info frame too please.

> There doesn't seem to be a fundamental flaw in the system: 

What an optimist!  ;-)

> I wonder if it is some thing in the way init runs the tasks in start_child.

The only thing really funny about it is that there are no fds.
You could try seeing if that breaks things in a chroot'd environment
(exec 0<&- 1>&- 2>&-; exec /libexec/console-run blah).

> However, I am not sure what can be tried next.
> Debugging console-run turns out to be hard.

I tend to blame gdb.  It always seems just a bit too finicky to me.