Incorrect time difference against a ghost replay about stk-code HOT 2 OPEN

After investigation, this has two main causes.

First, when crossing a lapline, it can happen that the replay recorder saves a distance that doesn't account for the crossed lap. This is the first cause.

138.641663 -105.644043 6.146392 67.657158 -0.005352 0.935016 0.020872 0.353949 26.120684 1.000000 0.252395 0.251139 0.246714 0.248570 1 0 0.000000 4 4 0 5224.465820 0 0 200 0 0
138.733337 -104.367973 6.179694 65.620842 0.000447 0.924822 0.003547 0.380384 26.427118 1.000000 0.243190 0.246226 0.248869 0.254281 1 0 0.000000 4 4 0 2613.485352 0 0 200 0 0
138.833328 -102.906494 6.185417 63.410870 0.003801 0.912611 -0.004915 0.408782 26.691500 1.000000 0.260125 0.260078 0.253631 0.258309 1 0 0.000000 4 4 0 5229.619141 0 0 200 0 0

The second issue is in how getTimeForDistance handles a ghost kart having been at a given distance multiple times. It starts its search for the relevant replay frame from the current replay frame (that is, from the position the ghost kart is at when racing against it), so if a ghost kart has been at a given distance multiple times, which of those times get used for the comparison depends on how quickly the other kart is driving.

Taking the earliest time the distance has been reached by the ghost would seem like a logical solution, but on the other hand it is clearly weird if the ghost kart is visibly behind and it is displayed ahead. It is rather good behavior that, if the ghost kart goes backwards and the kart it's compared to is now ahead on the track, its time is compared to the next time the ghost kart reaches this distance rather than the first one it did.

from stk-code.

First, when crossing a lapline, it can happen that the replay recorder saves a distance that doesn't account for the crossed lap. This is the first cause.

On testing, this happens on most laps in any kind of race. m_overall_distance gets updated accounting for the new position on track but before the new lap is validated, and is therefore wrong for one frame.

EDIT : New lap updates distances again, however, and further testing indicates that the replay recorder is only recording data afterwards. But it is a hard fact that the replay files contain the incorrect distance... Something is amiss.

from stk-code.