[OSRM-talk] Segfault with 5.4.0 osrm-extract and foot.lua

Daniel Hofmann hofmann at mapbox.com
Tue Oct 11 12:01:58 UTC 2016


For the record, tracking over at
https://github.com/Project-OSRM/osrm-backend/issues/3033

On Mon, Oct 10, 2016 at 6:27 PM, Michael Krasnyk <michael.krasnyk at gmail.com>
wrote:

> Hello Frederik,
>
> thank you for reporting and bisecting the first bad commit.
> I was able to reproduce the issue with foot profile for Europe and at the
> first glance the problem is in access to an stxxl vector name_data and/or
> name_offsets in CmpEdgeByInternalSourceTargetAndName call
> std::lexicographical_compare.
> I would suggest to create an issue at https://github.com/Project-OSR
> M/osrm-backend/issues to keep a track of it.
>
> From my side i will try to bisect name id during the week that leads to
> memory corruption.
>
> Kind regards,
> Michael
>
>
> On Mon, Oct 10, 2016 at 10:46 AM, Frederik Ramm <frederik at remote.org>
> wrote:
>
>> Hi,
>>
>> On 10/07/2016 09:43 AM, Frederik Ramm wrote:
>> > I have finished bisecting the commits and ended up with
>> > 2b466b2fb29c416149b4d881c151349218a63e1e as the first bad commit -
>> > everything before that works, everything after that segfaults. I'll have
>> > a closer look.
>>
>> I'm afraid I couldn't pinpoint the problem. I'm pretty sure there is
>> *some* kind of memory corruption but exactly when and where it hits
>> seems to be dependent on the STXXL configuration among other things.
>>
>> To summarize:
>>
>> * The problem is always reported as "double free or corruption
>> (fasttop)" and leads to a segfault in osrm-extract on Ubuntu 16.04.
>>
>> * The problem affects all OSRM versions from commit
>> 2b466b2fb29c416149b4d881c151349218a63e1e on (this means v5.2.7 is the
>> last-known-good release). I read through that commit and couldn't
>> immediately see anything that looks broken.
>>
>> * The problem does not occur with the "car" profile but it does occur
>> with the "bicycle" and "foot" profiles.
>>
>> * The problem does not occur with data extracts smaller than ~ 2 GB but
>> it does occur with larger ones. Sadly that made it impossible for me to
>> find a very small input file that would provoke the crash, and any
>> "valgrind" debugging was out of the question.
>>
>> I'll leave it at that and work with 5.2.7 which is good enough for my
>> use case. Given that you say all profiles work for you in 5.4, it might
>> be a funny edge case on Ubuntu or even on this particular machine I was
>> running it on. Time permitting I might re-run on a different machine
>> some time later.
>>
>> Bye
>> Frederik
>>
>> --
>> Frederik Ramm  ##  eMail frederik at remote.org  ##  N49°00'09" E008°23'33"
>>
>> _______________________________________________
>> OSRM-talk mailing list
>> OSRM-talk at openstreetmap.org
>> https://lists.openstreetmap.org/listinfo/osrm-talk
>>
>
>
> _______________________________________________
> OSRM-talk mailing list
> OSRM-talk at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/osrm-talk
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/osrm-talk/attachments/20161011/a2aecb70/attachment-0001.html>


More information about the OSRM-talk mailing list