Wind follower fx eventually breaks

[Drahsid]

This isn't really much of an issue since it takes like two weeks straight for this to happen, but here's the story:

• I did a small test of the game
• At some point I summoned the little whirlwind/tornado guy
• Left the game on
• I thought I was crazy, but it seemed like over the days that the spinning effect was very slowly speeding up
• Eventually, after a couple weeks, the effect stopped entirely

I'm guessing this is due to a timer accumulation error (underflow or float error?)

[lambertjamesd]

I use a 32 bit float to represent total game time. I'm guessing that it got big enough that adding 1/30 to it stopped making it bigger

[Drahsid]

Looking at the related code, that does seem to be the case. Some quick napkin math:

• Around 262144s or about 3 days, the step becomes 0.03125s
• Around 524288s or about 6 days, each step rounds to 0.0625s
• Around 1048576s or about 12 days, adding 1/30 has no effect

Fixing this particular issue seems pretty trivial, but I think the root cause is effecting a couple other things (for example

spellcraft/src/spell/mana_pool.c

Line 31 in 86c9671

float time_to_last_cast = game_time - pool->last_request_time;

)

It's good practice to use locally bounded floats which wrap or reset for this kind of thing, but it is also arguably better to use an unsigned integer for global timing since it's wrap safe, and also doesn't accumulate float rounding error and precision loss as time balloons. You would actually be surprised at how much error is introduced when accumulating floats, let alone doing it at extreme values.
I think mixing the two ideas (local floats + uint global time) would result in the best precision and also potentially reduce the memory throughput.

One more thing is that changing scene once the global timer has accumulated so high causes the game to crash with a generic float exception.