charleslemaux/Mirror
1
0
Fork 0
mirror of https://github.com/MirrorNetworking/Mirror.git synced 2024-11-17 18:40:33 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

docs: contribution criteria for optimizations

Browse Source 
Explain what an optimization pull request needs in order to get merged
This commit is contained in:
Paul Pacheco 2020-02-24 08:30:41 -06:00 committed by GitHub
parent 215e973903
commit ed9d056350
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 23 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
CONTRIBUTING.md
View File

@ -102,6 +102,29 @@ Submit your pull requests to the right branch:
If your pull request breaks any test, it has no hope of being merged.
#### Optimizations
There are generally 2 types of optimizations, micro-optimizations and macro-optimizations. The distinction has nothing to do with how much they improve the program.
Micro-optimizations try to improve the performance of an application by replacing instructions with equivalent but more efficient instructions. Some example micro-optimizations include:
* replace `i / 4` with `i >> 2`
* eliminate an allocation.
* replace `Vector3.Distance(a,b) < K` with `Vector3.SqrMagnitude(b - a) < K * K`
* convert a class to struct
Macro-optimizations try to improve the performance of an application by changing the algorithm. Some examples include:
* Serialize a message once O(1), instead of for every single client O(n)
* Change interest management algorithm, as of this writing every object checks every other object O(n^2), it could be replaced by a sweep and prune algorithm O(n)
* When synchronizing movement, instead of synchronizing every position change, you could synchronize the velocity and let the other side predict the position.
Macro-optimizations tend to change the **scalability** of mirror, by changing an algorithm, you may now support 10x more customers on the same hardware, it is even possible for a macro optimization to make performance worse for small numbers. Macro optimization usually make a really big difference, but are much harder to make.
Micro-optimizations tend to change the performance of mirror in a linear way. There are some micro optimizations that make a huge impact on performance such as eliminating allocations in the hot path.
We prefer readable code over optimal code. We do not like any kind of optimization if it makes the code less readable (they generally do). For that reason, we require that both micro and macro optimization pull request come with screenshots profiling a real game or at least a synthetic **representative** test. It is not enough to show that one operation is faster than the other, you must prove that this makes a significant difference in Mirror or in a real game using Mirror.
If your optimization pull request does not come with profiling data showing real gains in a meaningful test is has no hope of getting merged.
## Coding conventions
Start reading our code and you'll get the hang of it. We optimize for readability: