gamehour bug on OS X

Post by **Zini** » 25 Mar 2012, 10:11

Found a regression in 0.13: gamehour goes NaN if you move mouse in exteriors. Can anyone confirm?

Nope. Doesn't happen here.

corristo · Post by **corristo** » 25 Mar 2012, 10:35

Zini wrote:Nope. Doesn't happen here.

thanks. Will investigate.

corristo · Post by **corristo** » 25 Mar 2012, 14:41

I'm experiencing following bug on current master (0.13)
If I move mouse in game, gamehour eventually becomes NaN. If following line commented, everything is ok:

Code: Select all

hour = std::fmod (hour, 24);

(World::setHour).

So hour in some cases hour evaluates to NaN.
Any ideas how it can be related with mouse movement? Memory corruption somewhere?

Sky doesn't work properly because of this.

Didn't try to rebuild with gcc yet (using clang now), so it can be compiler issue. If anyone with clang can try, it would be awesome.

corristo · Post by **corristo** » 26 Mar 2012, 06:35

there is no bug in 0.12. Let git bisect help me!

Post by **Zini** » 26 Mar 2012, 10:15

Please do so. Note that this is the last bug that stops us from releasing 0.13.0. Don't wanna put any pressure on you

corristo · Post by **corristo** » 26 Mar 2012, 13:39

Updates:

with git bisect I found that commit. Previous commit 4c240c644fc2baad0b6def265fc4b715d094e39d have no bug. So I tried master with --no-sound — no bug. So it's somehow connected to sound.

werdanith · Post by **werdanith** » 26 Mar 2012, 13:50

To be more specific, when sounds can be played (music causes no issue), and then you move the camera, either by moving it with mouse or the keyboard then GameHour becomes NaN. And the first thought would be to blame the fact that the listener's position is updated when you move the camera, but that functionality was broken that far back in the commit tree, the listener's position was never updated. So this makes no sense at all ...

corristo · Post by **corristo** » 26 Mar 2012, 18:08

Looks like problem is deeper. Disabling sound only reduced it's appearance rate, but in some cells it still here =\ So I'm starting my time machine again and going to digg in times of New Sky Rendering Invention

It's some kind of Heisenbug

lgromanowski · Post by **lgromanowski** » 26 Mar 2012, 19:10

corristo wrote:Looks like problem is deeper. Disabling sound only reduced it's appearance rate, but in some cells it still here =\ So I'm starting my time machine again and going to digg in times of New Sky Rendering Invention

It's some kind of Heisenbug

Search for uninitialized variables/class members in code related to time calculations, probably that could be source of problem.

corristo · Post by **corristo** » 26 Mar 2012, 20:13

Let's blame clang. With good old gcc no problems.
I think we can release.
Will review/address clang issue later.

openmw.org