This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH] Fix use of a dangling pointer for Python breakpoint objects
- From: Pedro Alves <palves at redhat dot com>
- To: Pierre-Marie de Rodat <derodat at adacore dot com>, gdb-patches at sourceware dot org
- Date: Thu, 23 Jun 2016 17:14:47 +0100
- Subject: Re: [PATCH] Fix use of a dangling pointer for Python breakpoint objects
- Authentication-results: sourceware.org; auth=none
- References: <20160621104021 dot 15093-1-derodat at adacore dot com>
On 06/21/2016 11:40 AM, Pierre-Marie de Rodat wrote:
> Hello,
>
> When a Python script tries to create a breakpoint but fails to do so,
> gdb.Breakpoint.__init__ raises an exception and the breakpoint does not
> exist anymore in the Python interpreter. However, GDB still keeps a
> reference to the Python object to be used for a later hook, which is
> wrong.
Urgh, this code is ugly.
So the problem is that the next time gdbpy_breakpoint_created
is called, for some other breakpoint, we'll dereference the dangling
pointer then, correct?
> +
> +# Skip all tests if Python scripting is not enabled.
> +if { [skip_python_tests] } { continue }
> +
> +gdb_test "source py-breakpoint2.py"
> +
> +# The following used to trigger an internal error because of a dangling
> +# reference to a Python breakpoint object.
> +gdb_test "start"
"start" doesn't work with "target remote" testing. Try:
$ make check \
RUNTESTFLAGS="--target_board=native-gdbserver" \
TESTS="gdb.python/py-breakpoint2.exp"
Can we instead run to main first, and then source the python
script?
> +
> +bp1 = gdb.Breakpoint('main', gdb.BP_BREAKPOINT)
I don't understand the idea behind creating this breakpoint
before the failing watchpoint one.
> +
> +# The following will create a breakpoint whose construction will abort (there
> +# is no such symbol), so GDB should not keep a reference to the corresponding
> +# Python object.
> +try:
> + bp2 = gdb.Breakpoint('does_not_exist', gdb.BP_WATCHPOINT)
> +except RuntimeError:
> + pass
> +else:
> + assert False
Wouldn't it better to create a breakpoint after the one
that failed, explicitly? Either in python, or perhaps
simpler, a regular command line breakpoint directly in
the .exp file.
Ah, I think I see -- I guess the test is relying on "start" creating
a magic breakpoint at "main", and that one being the one
that dereferences the dangling pointer. But, see above about
remote testing.
Thanks,
Pedro Alves