Ribis

if you should want collisionchecking on same object

you can put the one object in a family and then check between the family and the object

or use system pick nth instance in a subevent (collision filters 2 instances)

sytem pick instance 0: do something

system pick instance 1: do something

and probably also by variable or uid, ..

—

[quote:9816sj4i]As for that little test of yours - drawing performance isn't what you should be testing here,

great, you explain your first condescending tone and start another one...

this was a test on drawcalls with multiple instances or multiple objects,

but please if you want to put the effort in it, add your test, it would be interesting.

Colludium , thanx for the test

so yes, using different objects causes a slight overhead, what i find interesting is that its not a constant cycle, draw calls go as low as 0.1 (same object) and then go back to maximum calls for a while, maybe thats garbage collection and then a redraw is needed, there is also not much difference in using canvas2d for me.

changing the object size and alpha values seems to take an extra hit, im not quite sure how drawcalls actually work.

optimizing only when needed, i understand, problem with that is, if you suddenly realize that you need extra performance, changing the way you setup things after doing all the logic is a real pita.

about testing this, it would be interesting.., most performance tests are just testing how much instances can be generated vs fps

but lets say generate 1000 instances, with one object and 1000 instances with 50 objects, i think i can test that pretty quick with nickname plugin, maybe i give that a shot later..

Ribis ,Colludium,Stiivais

while its true that its probably easier to have different object, most benefit is when you want to do a compare obj a to b / for collision, placement, and like you said to split things up for behaviors, etc.., and i was always a believer that this was better

but, i recently have doubts about what gives best performance, i read on the forum, that drawcalls are per object type, but what does that mean exactly?

does that mean that if everything is in one object there's only one drawcall? does that mean that everything in a family is one drawcall? im leaning toward thinking that one single object would have the best performance, plus you have the added benefit that single objects get spritesheeted, i currently have a good way to work like this, but its more for interface work.

first of all, its pretty strange you have better performance without webgl,

i would say its your graphic driver thats need updating, but then again if it happens on ios, that couldn't be the case

maybe its a genuine bug with the canvas render and tilemap, as most people are using webgl, it could be undetected

best would be to try to reproduce it in a smaller project and post a bug

above the fullscreen detail setting, "fullscreen in browser" has options, try integer scaling

Ribis are you using fullscreen scaling? you must use integer scaling, other than that, maybe post a capx with your settings to check

i thought about this problem, and its really bugging me, because ...

1. i want to use linear, it too nice to not use it

2. its perfect on preview

3. other programs use it

4. i want to use construct

consider this linear example

a 16 at 16 tilemap, about 76 frames

sprite tilemap imported from tiled (no plugins needed) hold mouse to move, mousewheel to zoom, starts at zoom 2

download and preview file / perfect no seams no bleeding, even when scaling / construct can

https://dl.dropboxusercontent.com/u/61666915/tilemaptest.capx

on export this happens, the actual background is showing thru = seams, ( no bleeding tho ) / construct can't

https://dl.dropboxusercontent.com/u/61666915/tilemaptestv2/index.html

so as showed in my early version with the colors (modified export), its possible to have tiles perfectly aligned and no seams, so this means, whatever happens it is happening internally on the texturemap and not on the positioning of the tiles

ps: what you will have and always have is a little bit of blur from the surounding color but this is not an issue and almost not noticeable (well check the preview)

so its pretty clear to me that construct engine can display linear perfect, so its really a question what happens on export

for the bleeding there's the 1 pixel transparent gap, not actually a problem, but it needs adjustments, it must consider transparent space for two tiles and a setting for these transparent pixels

libgdx has its own texturepacker, there are the follow options, and i think the bleed one is actually another name for what extruding actually is

these are there standard settings:

paddingX The number of pixels between packed images on the x-axis. 2

paddingY The number of pixels between packed images on the y-axis. 2

bleed If true, RGB values for transparent pixels are set based on the RGB values of the nearest non-transparent pixels. This prevents filtering artifacts when RGB values are sampled for transparent pixels. true

edgePadding If true, half of the paddingX and paddingY will be used around the edges of the packed texture.

link: https://code.google.com/p/libgdx/wiki/TexturePacker

this makes sense, because linear will pick the surrounding pixels of the texture actually the subtexture, to make transition, and now the two tiles takes the transparent pixels of the 1px gap, that is why we see this obvious border around the tiles on export.

theres also an analogy between 3d models used in games, always padding between uv's and never stop your texture on the uv-border or you get the same black lines on your model

now for the c2 tilemap, if it would use the same texture method, it could still use the optimization, because again, the texture size is not changed, and there is no overlap.

so i hope you consider testing these settings out, i don't know if this is a convenient workaround, or easy coded, but its definitly worth trying something other engines are putting in and something where most of us are having trouble with, especially when you consider that on preview you could asume that everything is working fine...

using a smaller layout and set fullscreen to low quality (important) should improve it alot, as you don't have much logic going on, its most likely already limited by gpu

in cocoonjs, simply updating a text/textbox every tick (for fps display for example) will reduce framerate, so always use spritefont

crosswalk renders textboxes better because its a real browser, but again spritefont is better

i'm not sure about layers actually, i try to limit my usage as much as i can, but i haven't done much testing on the difference

only way to have it pixel perfect (original resolution) is not use scaling

have you tried how it looks with fullscreen cropmode? you could add grass for empty space, or a scrollmode for lower resolution screens

adding on that, you could test then how a layoutscale of 2* looks

other than that integer scale should be the best option

thanks!, tested and working