The glitch here is the double assign in a single frame.
Yeah, either support it fully or disable it completely.
We were able to desync physics and replay even without same-frame assignments:
- pause
- assign miner to lem A
-- Expected and observed: physics advance, sound plays, A becomes miner
- press [
B] to go back in time by 1 frame
- assign miner to lem B different from lem A
-- Expected: Either double assignment (which you don't want) *or*
the now-pending assignment to A is cancelled from replay and physics,
then physics advance, and B will become miner
-- Observed instead: physics advance, no assign sound plays at all,
assignment to B is ignored completely by the physics,
and A becomes miner again
- assign miner to lem B again
-- Expected and observed: physics advance, sound plays, B becomes miner.
A continues to mine. B's animation is one frame behind that of A,
because we have assigned to B one frame later than to A.
- press [
pad-minus] to go back a second, to a point in time way
earlier than A's assignment
- unpause and watch our recording
-- Expected: First A, then B get an assignment
-- Observed instead: only the assignment to B is replayed
There is a variant of this.
- pause
- assign miner to lem A
- press [
B] to go back in time by 1 frame
- assign miner to lem B different from lem A
-- same as above so far, same discrepancy in expected/observed behavior
- press [
pad-minus] to go back a second, before A's assignment
- unpause and watch our recording
-- Expected: assignment to A at the proper point in time
-- Observed instead: no assignment to A ever happens
Or even:
- pause
- assign miner to lem A
- press [
B] to go back in time by 1 frame
- assign miner to lem B different from lem A
-- same as above so far, same discrepancy in expected/observed behavior
- unpause, watch A mine for 5 seconds
- press [
pad-minus] to go back a second
-- Expected: A is still miner, and the miner tunnel got shorter than it
was before we pressed [
pad-minus].
-- Observed instead: A didn't get any assignment 4 seconds ago,
and is a walker. By now, he's walked somewhere completely different.
This is a
source of erroneous replay files without being able to notice they're erroneous.
Players can produce a replay that misses some assignments, but don't notice it, because they never hit [
B]/[
R]/[
pad-minus]. The physics don't match what will be saved to file.
-- Simon
<SimonN> Akseli: do you remember whether this [bug described in this post]
was already in vanilla Lemmix?
<Akseli> I don't think so, didn't encounter this in Lemmix
<SimonN> hmm, it might be worthwhile to add this, so he can better guess where the bug was introduced