Re: Improved linker-debugger interface

[Tom Tromey <tromey@redhat.com>]

Paul> 1. Stripped binaries.

Paul>    There is a DT_DEBUG entry pointing to _dl_debug_state (set by ld-linux)

Paul>    You might want to have a new dynamic tag, pointing to
Paul>    _dl_debug_state_extended(), so the debugger would be able to track
Paul>    shared libs using the new mechanism even when everything is stripped.

SystemTap probes aren't stripped, so I think this should work fine.
This is one of the big plusses (from my POV) of using probes, and it is
why we introduced probes into longjmp and the exception unwinder too --
it lets gdb features work even when the user hasn't installed debuginfo
for glibc and/or libgcc.

This stuff is in F15.  It wasn't accepted into glibc mainline.

Paul> 2. In-process debuggers.

Paul>    B) compiles down to a single 'ret', so there is no way to place a patch
Paul>    on top of it.

Right now a SystemTap probe is a single nop, so probably no joy there.
We have talked about making it a wider nop, though, to facilitate this
kind of patching, primarily I think by LTTng.  What do you think of
this?

Paul>    Perhaps co-operation with "in-process" debuggers would be more
Paul>    acceptable for a new interface?

It is fine by me.

Tom