Xenial's gdb 7.11 breaks rr
me at kylehuey.com
Sat May 14 00:23:21 UTC 2016
On Fri, May 13, 2016 at 5:19 PM, Kyle Huey <me at kylehuey.com> wrote:
> On Fri, May 13, 2016 at 5:14 PM, Robert O'Callahan <robert at ocallahan.org>
>> On Sat, May 14, 2016 at 12:09 PM, Kyle Huey <me at kylehuey.com> wrote:
>>> 1. Build tip rr (8566e3038892)
>>> 2. ./bin/rr record ./bin/reverse_continue_process_signal
>>> 3. ./bin/rr replay
>>> 4. c
>>> 5. Once, the program has reached the final SIGKILL, b 22
>>> 6. rc
>>> Broken versions of gdb appear to forget which direction execution is
>>> heading in when an intermediate (and uninteresting) signal is reported by
>>> rr to gdb and resume with a vCont instead of a bc which sends us back to
>>> the SIGKILL at the end.
>>> This is fixed on the gdb tip, but building gdb is sufficiently annoying,
>>> and there is enough intermediate breakage, that I did not pin down a fix
>>> range. Downgrading gdb on Ubuntu is non-trivial as well, so I ended up
>>> having to build from source :/
>>> I encountered this debugging Firefox, where the SIGKILL was the shutdown
>>> hang timeout killer and the intermediate signal is a SIGCHLD.
>> Yeah, and there's also
>> https://sourceware.org/bugzilla/show_bug.cgi?id=19855 :-(.
>> Can we get the fix for your bug backported to 7.11.1? And maybe see if
>> Ubuntu will update gdb to 7.11.1 in Xenial?
> I would have to identify what actually fixed it first.
Lest I sound too pessimistic here, I'm willing to do the work if getting it
uplifted seems plausible. Do the gdb maintainers ship dot releases? Does
Ubuntu take them?
For 19855 I think we'll need to have rr sniff the gdb version somehow and
>> stop doing 'maint target-async 0'. That'd let us avoid the hideous
>> workaround for https://sourceware.org/bugzilla/show_bug.cgi?id=18617 as
> - Kyle
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the rr-dev