[CrystaX.NET][multi_index][release] Segfaults with Clang 3.5 on Android
Hi, Android CrystaX.NET-* testers are segfaulting on some (~65%) of the Boost.MultiIndex tests after successful build. For instance: http://tinyurl.com/jwnfryy The curious thing is that this is Clang 3.5 only, only for some tests and the same tests across the different Android variants. I'd like to help fixing whatever it's required (if anything) in the lib so as to make this work, but the existing info does not provide any clue. Maybe the regression tester maintainer can shed some more light here? Thank you, Joaquín M López Muñoz Telefónica
I'd like to help fixing whatever it's required (if anything) in the lib so as to make this work, but the existing info does not provide any clue. Maybe the regression tester maintainer can shed some more light here?
Reason of segfaults is not clear for me too. Call stack is very short, having boost/multi_index_container.hpp:558 as top-level entry - for all crashes. This makes me think call stack is corrupted for some reason. Also, taking into account all crashes are clang-3.5 only and passed for clang-3.4, I suspect it could be bug in clang-3.5 toolchain in our NDK. This is not the first time I see such bugs in clang, so it probable. I don't really think you could fix these crashes on your own, unless you take CrystaX NDK and start debugging it on low level, fixing&rebuilding clang if needed. Perhaps it would be more reasonable to let us debug this issue since we're much more in context of Android specifics - and for me it looks much more probable that bug is in clang, than in multi_index code. I've filed ticket https://tracker.crystax.net/issues/908 - so you can subscribe there to easily track status changes. -- Dmitry Moskalchuk
Dmitry Moskalchuk <dm <at> crystax.net> writes:
I'd like to help fixing whatever it's required (if anything) in the lib so as to make this work, but the existing info does not provide any clue. Maybe the regression tester maintainer can shed some more light here?
I don't really think you could fix these crashes on your own [...] and for me it looks much more probable that bug is in clang, than in multi_index code.
I've filed ticket https://tracker.crystax.net/issues/908 - so you can subscribe there to easily track status changes.
Wonderful, thank you very much. I'll be following the ticket in case contributions from my side are needed. Joaquín M López Muñoz Telefónica
Dmitry Moskalchuk <dm <at> crystax.net> writes:
I'd like to help fixing whatever it's required (if anything) in the lib so as to make this work, but the existing info does not provide any clue. Maybe the regression tester maintainer can shed some more light here?
[...]
I've filed ticket https://tracker.crystax.net/issues/908 - so you can subscribe there to easily track status changes.
Hi Dmitry, Well, the bug was in the lib itself, not the environment: http://tinyurl.com/k3bfb99 The thing was an null-dereferencing UB issue that every compiler has been (fortunately?) skipping right for years till this one. Thank you for your help and sorry for the noise. Joaquín M López Muñoz Telefónica
On 01/05/15 11:18, Joaquin M Lopez Munoz wrote
Well, the bug was in the lib itself, not the environment:
The thing was an null-dereferencing UB issue that every compiler has been (fortunately?) skipping right for years till this one. Thank you for your help and sorry for the noise.
Thank you for letting me know. I've closed ticket https://tracker.crystax.net/issues/908 as invalid. -- Dmitry Moskalchuk
participants (2)
-
Dmitry Moskalchuk -
Joaquin M Lopez Munoz