With the old code, if the object at the head of the list
was removed, it would leave the reference behind,
extending the lifetime of the thinker until P_RunKartItems
or P_RunOverlays was run.
This fixes some thinkers never being removed due to having
negative reference counts.
And here's a breakdown of why the old code could produce
negative reference counts:
Consider P_CheckPosition. This function calls P_SetTarget
on tm.thing but does not call P_RestoreTMStruct. This
means that tm.thing will not be NULL the next P_SetTarget
is called on it. What are the implications of this?
Consider the following series of events:
- P_CheckPosition is called, tm.thing != NULL afterward
- Another function saves the tm struct and sets tm.thing to a different mobj
- - the old tm.thing will have its references decremented
- - the new tm.thing will have its references incremented
- This function calls P_RestoreTMStruct
What should happen when P_RestoreTMStruct is called? The
*new* tm.thing should have its references *decremented*
and the *old* tm.thing should its references
*incremented*, of course, for symmetry with P_SetTarget.
The old code correctly decremented new tm.thing's
references but did not increment old tm.thing's
references.
- If no clients at server start or after 10 seconds of GS_LEVEL, and no Netxcmd waiting to be digested, halt all SV_MakeTic.
- It's absolutely netsafe to only have enabled on the host's end, the only risk is that a dedicated server might not re-awaken when presented with certain stimuli
- Fix allocating only enough for "MAPXX".
- Fix memory leak in M_StartTimeAttack.
- Use va instead of malloc + sprintf, strdup where needed.
Some of these pass gpath to va anyway so that should be
enough room.
- Players moving into a gate with a lot of momentum would
offset the origin point from its intended position
- Find intersection of player movement and gate
- Correct origin point between loop entry and exit by
interpolating the difference between player position on
entry and intersection point
SRB2 uses a LOT of mapthing args compared to Hexen (which has none) and ZDoom (which only has them on objects that will never ever activate scripts). So we really badly needed to separate the two if we want attaching scripts to things to be useful.
A first pass in attempts to fix crashes when ACS causes map changes.
Frustratingly still got a crash, but I think this is definitely the right foundation to work with.
