Concerns from a "Serious" developer

Post » Mon Apr 03, 2017 6:51 pm

digitalsoapbox wrote:
Jayjay wrote:
Ashley wrote:Do you want us to build native engines? I've covered that in this blog with our rationale around that, which nobody ever really directly argues against, there's just vague accusations of how HTML5 is "poorly optimised" or something, which really is not the case given the potency of modern JIT compilers and the native-equivalent performance of WebGL.


Actually I and others do have a problem with the arguments in that article, as well as the article on "HTML5 games faster than native"

Let's see:

Ashley wrote:WebGL is close enough to OpenGL to be considered the same performance-wise. In other words, Construct 2 games render in practically the same manner a native engine would.

...

We already cover Xbox One and Wii U - so that only leaves PS4. Currently we're blocked by Sony not providing an easy way to publish HTML5 games to the PS4. Given how well our Xbox One support worked - requiring no significant changes to our engine to work on a console - we are keen to see Sony supporting HTML5 on their console too.


There's a few arguments against those lines of thought:

Jayjay wrote:Listing GPU fillrate as the main cause of a games slowdown... and then stating that the WiiU is a viable export platform is quite funny. WiiU doesn't support WebGL, meaning not only regular GPU "fillrate issues", but no special FX and low performance.

5FPS after losing WebGL and FX is still not playable for a platformer. We didn't even make an HD title, just a retro 16bit platformer with 32x32 to 64x64 pixel sprites on average.

As for fillrate being "the main culprit", there are many instances where CPU should be considered too (eg: most enemies and players using the platformer behavior) and, no matter the optimizations, aren't going to be able to combat the issues of JavaScript without emscripten. But, if you make a tool that exports to HTML5 through emscripten it might as well also export the actual C/Cpp (to port to the JS, best of both worlds).


RafaelHp wrote:Even the mobile games developed with Unity and Unreal put C2 at the ground... How you can even compare?

- Native games is WAY better in lower and mid hardwares, and this is AMAZING because more people will play your games.
->>Scirra depending on thirdy-partys for realese our game. Remember when you guys drop support to Cocoon as exporter? That's why I droped C2.

...

By the way, The Next Penelope is being ported to C ...


Gianmichele wrote:To me the main issue I had with html5 games is that on low-end devices (iPad 2) they perform not really well at all. The same simple app on iOS, even with the super accelerated safari, tends to have small hickups and feels not smooth compared to the same done in a native engine (love2d in this case).
The situation seems a bit better on newer devices (iPhone6), but still showing those small jerks every now and then (are they GC related?).

All of this scares me away from trying to do something serious with html5 engine.


R0J0hound wrote:While it's true most devices support webgl it doesn't mean we get the same performance as native in that respect. In my experience it varies per device, and it's just not about the hardware capabilities. Case in point, for me in CC rendering is way faster than in C2. This goes against your benchmarks so that leads me to believe webgl gives native performance on only certain devices. The argument that the hardware is the limit doesn't fly when changing software and doing the same thing performs a lot better.

The browser's hardware support is primarily about having everything work instead of having everything have good performance imo. Even things like the vsync jank is still there on some hardware. This is probably what causes the disparity between the issues users see and what is tested.

I don't know if native support would fix this, and I realize doing so would cause a load of different issues and multiple code bases, but there seem to be core issues with how browsers do things that are not performant.


Now, in regards to "HTML5 faster than Native?":

Ashley wrote:Construct 2's WebGL renderer is faster than our previous native C++ DirectX 9 based engine - the one we developed for Construct Classic.


This is a completely wrong comparison, because C2 is largely an improvement and optimization on CC, you already learned from past mistakes and found new ways to optimize. It's the software form of a "straw man" argument against native.

If you back-ported those optimizations back to C++ you'd find that the results suddenly return to native being the fastest and smoothest. We can tell this ourselves from simply our experience of porting to Unity in C# with our current game and seeing it run much faster and with more lighting, special effects, and enemies on-screen than our C2 prototype.

Then there's also the people who didn't see the results that are promised in that article with C2 vs CC:

dEspadas wrote:97151 - WebGL
128458 - Native
i3 - 3220 / GeForce GTX 550ti with small factory overclock / Win 7 64 / nvidia driver 306.97 / 8Gb RAM


Epox wrote:Intel E2180 2Ghz, 4GB RAM, Win7 64 bit, SP1, Geforce 210, Nvidia 306.97 driver

Native/DX 25613

WebGL:
Chrome 22: 3801
Chrome 23: 4482
Chrome 24: 3901 (dev)
Chrome 25: 3503 (chanary)
FF16: 3254


Rayek wrote:i7 9203.6ghz, 48gb, win7 64bit , amd7970 3gb, catalyst 12.11(beta)

Native/dx: 118040

WebGL:
Firefox 16.0.2: 58712
Chrome 22: 103761


BruestleMulti wrote:WebGL Performance (Google Chrome 23): about 10400 objects
Native Performance: about 24000 objects

My notebook: Processor Intel Core i5-2410M 2.3 Ghz, 8 GB Ram, Graphic card Intel HD Graphics 3000, OS Windows 7 Ultimate 64-Bits


I'd like to point out two things from all of this:

1.) Like the data above, some of the "still-Steam-average" hardware still sees native far out-performing WebGL, and desktop is currently the best supported platform for HTML5 + WebGL. The results may be 4 years old, but the average 2D gamer (based on customer support issues on Steam) hasn't updated their PC/laptop/even just the graphics card in a long time.

Even with an Intel i7 6700k and GTX 1070 which has much better WebGL support than most of my customers I get:

Sprites/objects @ >= 60fps with no drops below
Native/dx: 86k
Chrome: 95k

This may not be the number of sprites that will appear in a game, but it still reflects an underlying performance gap between Construct Classic (which could be argued to be almost as un-optimized as C2 is optimized) and WebGL. HTML5 is not faster than native. It can get close with WebGL + Emscripten, but it inherently is just not capable of being as fast as native, and I wish that Scirra would stop trying to compare the two this way.

2.) You can't just ignore comments/evidence/arguments against your company/public statements about your engine and then claim "nobody ever really directly argues against" them!

Bonus: That multiplayer plugin would probably have been used a lot more if people were able to export their games and "publish everywhere". It's also possible the users are waiting for more helper functions/plugins/behaviours that make tracking objects much easier (Eg: a behaviour for syncing objects). I know multiplayer is generally a hard feature, but it's a great one to have even if it doesn't get used as often as other plugins.

Like others are saying, yes, export matters 100% more, but I still am glad Construct got multiplayer.

Although, it seems some other users had a better time using their own multiplayer solutions:

scoterr wrote:Using Electron instead of nw.js made my games smaller and better working, using my own networking solution with central socket.io based server destroyed weird, random connection problems that disallowed me from even testing my game. And made multiplayer a lot smoother. Being able to write my own rendering code allowed me for much more control over how stuff is shown on screen what makes performance better.


+1 to all of this, especially the "we're just as fast/faster than native" claims which are so completely beyond the pale in terms of just outright fabrications of HTML5's current abilities, using C2 or not. This is something I knew going in to C2, and I understand the performance difference (though it is greater than it should be with C2), so when I see this claim pushed over and over despite gallons of real-world evidence otherwise - and sorry, but your internal tests with tiny little games or demos mean nothing when compared to thousands of copies of commercial games out in the real world that show the opposite, and PLEASE don't try to blame the game's developers for this as you've done in the past - it's very hard to swallow after the 100th time when all evidence points to the opposite being true.


+1
B
6
S
1
Posts: 13
Reputation: 376

Post » Mon Apr 03, 2017 6:52 pm

Events can be abused way easier than code, and even the templates that come with C2 aren't designed for performance necessarily.

There's dozens of different ways to do things, and many of those are just terrible for performance.

I'm telling you there needs to be some exposition into the proper use of them.

You make a "for each object" that runs every tick, then the exporter should give a warning, etc.
Image ImageImage
B
172
S
50
G
183
Posts: 8,440
Reputation: 115,599

Post » Mon Apr 03, 2017 6:56 pm

newt wrote:Events can be abused way easier than code, and even the templates that come with C2 aren't designed for performance necessarily.

There's dozens of different ways to do things, and many of those are just terrible for performance.

I'm telling you there needs to be some exposition into the proper use of them.

You make a "for each object" that runs every tick, then the exporter should give a warning, etc.


Most of us that have used Construct 2 for a long time and have done the leg work already know this and optimize our games, and avoid poor event practices--so this has nothing to do with this post.
B
6
S
1
Posts: 13
Reputation: 376

Post » Mon Apr 03, 2017 6:57 pm

newt wrote:Events can be abused way easier than code, and even the templates that come with C2 aren't designed for performance necessarily.

There's dozens of different ways to do things, and many of those are just terrible for performance.

I'm telling you there needs to be some exposition into the proper use of them.

You make a "for each object" that runs every tick, then the exporter should give a warning, etc.



Have to agree with that being a useful feature, plus it means that instead of users finding work-arounds to issues like families (eg: often needing a for each object to actually let each object in a family do what it's supposed to individually) then the users would be forced to submit bug reports to issues and improve Construct for all.

Although I also agree with @jayrp1 because the users here are probably seasoned in the tool and have done whatever it takes to save performance already.
Construct Classic - Examples Kit Dropbox is a pile of trash and if you need my old files PM me! :)
B
127
S
43
G
18
Posts: 2,240
Reputation: 20,592

Post » Mon Apr 03, 2017 6:59 pm

Wow, a 22 pages long topic. While it was an interesting read, I have to admit that after 12 pages or so I had to just scroll though the pages, since it takes ages for me to read all posts through (hats of to those who do that). But one think caught my eye and I'd like to tag @Ashley for I think it's a great suggestion:
Thndr wrote:A page with Scirra tested benchmarks for each platform(PC/Mac/Linux with PC specs, Android/iOS with model, wrapper vs browser) with a Construct 2/3 version of the exported test file would work wonders in giving people an idea of how relative their current devices are to the test devices. If the benchmark demos even had an option to "submit your results to Scirra!" on it that would help aggregate consumer side data for a wider range of devices than the ones you can afford (but I do recommend having an official list of test devices that you can have physically)
B
137
S
33
G
17
Posts: 1,560
Reputation: 20,797

Post » Mon Apr 03, 2017 7:05 pm

Elliott wrote:
1) What exporters are missing from C2? They all work. Some work better than others; none of them are broken.



would it be a stretch to say that you haven't published or tried to publish a commercial game using c2? Because seriously, if you did, you'd know exactly what we're talking about.
We've spoken up about what the issues are plenty of times. Yes, the exporters "work." But that's it. That's like saying Photoshop does export PNGs, yet the image quality is low and the image is blurry... It DID produce a PNG though...

When we are here trying to get better exporters done, it's not going against anyone else's wants or needs so I'm not sure why there's so much backlash for asking for the exporters to work properly. We want our 2d games to work. They're 2d! I don't think it's too much to ask them to run as intended.
Last edited by NotionGames on Mon Apr 03, 2017 7:09 pm, edited 1 time in total.
Image
B
75
S
22
G
9
Posts: 572
Reputation: 14,171

Post » Mon Apr 03, 2017 7:08 pm

To be fair, since my post that jayjay quoted me with I have since got a replacement for my 10 year old computer so the graphics are no longer a bottleneck on my system. Also the few times I tried mobile export the performance was much better than my old system.
B
94
S
33
G
118
Posts: 5,394
Reputation: 75,849

Post » Mon Apr 03, 2017 7:10 pm

That would be great if Construct would use Electron instead of NW.js
And would add support to circular shaped objects (Shadow Cast and Collision) as well as fix the bug that would occur when the light is close to the object casting infinite shadow.
B
45
S
19
G
3
Posts: 102
Reputation: 4,598

Post » Mon Apr 03, 2017 7:11 pm

NotionGames wrote:
Elliott wrote:
1) What exporters are missing from C2? They all work. Some work better than others; none of them are broken.



would it be a stretch to say that you haven't published or tried to publish a commercial game using c2? Because seriously, if you did, you'd know exactly what we're talking about.
We've spoken up about what the issues are plenty of times. Yes, the exporters "work." But that's it. That's like saying Photoshop does export PNGs, yet the image quality is low and the image is blurry... It DID produce a PNG though...

When we are here trying to get better exporters done, it's not going against anyone else's wants or needs so I'm not sure why there's so much backlash for asking for the exporters to work properly. We want our 2d games to work. They're 2d! I don't think it's too much to ask them to run as intended.


+1
Excellent analogy.
B
6
S
1
Posts: 13
Reputation: 376

Post » Mon Apr 03, 2017 7:15 pm

I was appealing to the letter of the law, as Lamar seems to be working on a false advertising angle.

I don't doubt C2 is immensely frustrating to use for console games; I use it for web projects and mobile games because I believe that is what the software is truly suited for.

To continue with your image editor example, Paint can produce the Mona Lisa, it's not inaccurate to claim this - yet if I was to try, I'd rather use Photoshop, because that tool is correctly suited for the job.

If I was to make a console game, I would personally use Unity, because I feel that HTML5 is not the right fit for consoles yet; and won't be until browser technology is considered as important as native - which is coming, slowly but surely.

I honestly believe the exporters work perfectly; it's the support for the underlying technology that is flawed.

Now where I do have immense sympathy for you is how Scirra advertise their export support. C2 can theoretically make websites; yet they (rightly) don't advertise it as a web editor - could one argue that this is similar to the WiiU? Perhaps.
B
64
S
25
G
10
Posts: 654
Reputation: 10,842

PreviousNext

Return to General Discussion

Who is online

Users browsing this forum: No registered users and 0 guests