That's sane

I split this in to multiple smaller patches. a la D4798

In D4834#206091, @Stan wrote:

Did you look at the generated assembly, or is it just a guess?

No, it's a guess and I also guess it does not has much evect since components do only subscriped once.

This patch also allows components to subscribe other components to messages. Is this wanted?

No, I think it is acceptable. IMO ClassInit should be replace by an array of MessageTypeIds and LoadComponentTypes would lookup thous. But more redesigneing is required and that is not the purpose of this patch.

In D4730#206025, @jprahman wrote:

Lgtm, could probably bikeshed more on the existing futures code, but let's move this along :)

@mbusy The feature freeze will start on 3. Jan. I (and many others) hope this will be commited until then.

In D4830#205966, @vladislavbelov wrote:

In D4830#205964, @phosit wrote:

Use initialization over assignement. -> more const.

I'd like to, but then it adds implicit relation for the order of arguments. So I'd prefer to leave as is until C++20 designated initializers.

I thaught about this a bit more:

1. The standart does differentiate between allocator and memory resource. We schould follow that pattern to.
2. I think the reason for the performance improvement is that all components are stored together -> less cache misses. To address this it would be bether(even more perf improvement and less error prone) to store the components in the container instead of a pointer.

Use initialization over assignement. -> more const.

Yes that is much cleaner :)

copyright-year

Now the function name is not the best. Something like GetSingleChild or GetOnlyChild.

In D4830#205684, @vladislavbelov wrote:

In D4830#205677, @Stan wrote:

Yeah that's what I meant :)

Ok, will do. I think @phosit meant the same.

can you upgrade glad seperatly?

move to test

time the sleep duration there will be less fluctuation in the second test

I made a benchmark: D4829
Before the patch:

Benchmark:
async cost:
enqueueing:     377ms
work done:      377ms
total:  384ms
async round-trip:
total:  3809ms

And after the patch:

Benchmark:
async cost:     
enqueueing:     360ms
work done:      360ms
total:  363ms
async round-trip:
total:  4088ms

LGTM
Silhouettes are as buggy as before :)

Put variables in a struct

Copyright Year

more unique_ptr / my own comments

The particle work again.
can you add context.

Make it independent from the optional patch.

Implement a Find algorithm with projection. To avoid duplicate.

It works.
I also tested the shiny mod.

use this-> more consistently

include order

Changes to the queue and the constant.

fix apple clang

The game looks the same and I didn't notice a performance difference.

Move the initialization back

@vladislavbelov is it ok if i implement ranges::find? It wold be neat if i can use projection.

Implement Vlads aproach and use inheratance (empty base optimization) again.

In many files #include "renderer/WaterRenderer.h" is not required.

Red and Auburn are not easy to distinguish.

In D4812#204900, @wraitii wrote:

Could you rebased D4730 to only contain its own changes? Obviously the patch won't apply, but it'll make it easier to understand why e.g. this patch is necessary

smaller patch, now only the std::function in the TaskManager is replaced.
rebased to D4803. (it will not compile alone)

It is a clean up, since we now can use std::optional.
In the void case the optional needs some memory. Thats the only negative pervormance impact i can imagine.

Copyringht year and comment

The rest looks good and works

Applied comments from Vladislav

Generaly more things can be const.

I stay with this version with the 3 std::finds. It is to complex for to litle benefit.
I'll deduplicate the predicate.

In D4808#204716, @Stan wrote:

Maybe we could use a function where we pass CTexture::** to reduce duplication?

Yes, but...

EDIT maybe we could handle all cases in one iteration too not sure.

e.g sort states, look for HIGH_NEEDS_CONVERTING then PREFETCH_NEEDS_LOADING then PREFETCH_NEEDS_CONVERTING also check if converter busy to filter out

• on Echo Valley sometimes the cc's are covered in trees:
  
• there is no preview picture. It's not in the map browser.
• what does the team placement option do? I don't notice an effect.
• probably i'm a bit late for this comment: It would look less artificial if it would be rotational symetry instead of miror symetry. Also a odd number of players currently doubles the repetition: 7 players -> 14 repetitions.

In D4805#204608, @Stan wrote:

Is #5614 related to this?

Call UpdatePages before iteration. This should fix the test.

Applied (most) comments from Vladislav

In D4751#203653, @Freagarach wrote:

I guess I'll wait then. ^^

Add context