Make single link node pool unlimited

[TheNormalnij]

Fixes crash with offset 0x00133606.
The crash related to the high density of buildings.

[tederis]

You did a lot of work and it's great. I just want to understand why did you choose creating CDynamicPool instead of just using std::vector? In the context of this PR std::vector allows to find elements with O(1) complexity(by comparing data pointers), it also allows you to remove elements with O(1) complexity(swap and pop ideom). std::vector has to be a reasonable choice here.

Okay, I see that memory pieces have to be persistent and we cannot use swap and pop ideom here.

[Dutchman101]

Fixes crash with offset 0x00133606. The crash related to the high density of buildings.

This crash is exceedingly rare on MTA (measured by the latest build with a high spread; crash stats) so to judge the PR's feasability it would be good to know if it's worth it, as in: are there no negative/side effects at play, like a global increase of RAM usage as the pool allocates more? Is it dynamically expanding and going back down depending on a servers' usage of map and buildings, so that as long the usage isn't huge, there are no negative effects?

In the latter case it wouldn't be too bad to have it in. If it has no negative effects, proportionality to what it aims to fix isn't a thing.

@TheNormalnij If you could please make a list of any effects you expect from this change?

[TheNormalnij]

Fixes crash with offset 0x00133606. The crash related to the high density of buildings.

This crash is exceedingly rare on MTA (measured by the latest build with a high spread; crash stats) so to judge the PR's feasability it would be good to know if it's worth it, as in: are there no negative/side effects at play, like a global increase of RAM usage as the pool allocates more? Is it dynamically expanding and going back down depending on a servers' usage of map and buildings, so that as long the usage isn't huge, there are no negative effects?

In the latter case it wouldn't be too bad to have it in. If it has no negative effects, proportionality to what it aims to fix isn't a thing.

@TheNormalnij If you could please make a list of any effects you expect from this change?

This crash is rare because developers use createObjects instead of createBuilding when their map crashes.
I'm trying to increase the building limit because buildings fix fps issues on big maps. This allows more creative maps with better lighting.
This PR doesn't cause significant memory usage. I'll implement TEDERIS's suggestion, which requires an additional ~200 KB on game startup. createBuilding is still more memory efficient. 10k buildings use 1.5 MB less than 10k objects. The pool resets capacity on reconnect and can be resized manually via engineSetPoolCapacity.

The pool implementation is very universal. I plan to use it in future fixes too.

[Pirulax]

I feel like that a traditional linked list of these "part"s would be better than having a vector of them, especially when allocating a free-list is much better in terms of performance than a basically O(N) lookup.

[TheNormalnij]

I feel like that a traditional linked list of these "part"s would be better than having a vector of them, especially when allocating a free-list is much better in terms of performance than a basically O(N) lookup.

The performance cost for allocating m_poolParts isn't really important here. But I agree std::vector<std::unique_ptr<CDynamicPoolPart<PoolObjT>>> is weird. It should be std::list<CDynamicPoolPart<PoolObjT>> or std::vector<DynamicPoolPart<PoolObjT>>. The second variant causes less memory fragmentation.

[tederis]

The std::vector was fine because an array of CDynamicPoolPart ptrs isn't something frequently changing. Technically std::list is also okay, but for small types like uint or std::unique_ptr it performs worse because of cache misses and memory fragmentation. In most cases std::vector is the best choice when you need to iterater over an array of small fixed(non-volatile) objects. But nvm, I just wanted to note that your previous solution was also fine.

[Dutchman101]

Note for future investigation: This may have introduced a crash, see https://i.imgur.com/Or8lDDi.png for details
cc @tederis

I couldn't gather much from a quick look - but the crash occurs in this code.