#jogamp @ irc.freenode.net - 20150223 05:05:34 (UTC)
20150223 05:05:34 -jogamp- Previous @ http://jogamp.org/log/irc/jogamp_20150222050533.html
20150223 05:05:34 -jogamp- This channel is logged @ http://jogamp.org/log/irc/jogamp_20150223050534.html
20150223 08:13:27 * monsieur_max (~maxime@anon) has joined #jogamp
20150223 08:26:46 * eclesia (~husky@anon) has joined #jogamp
20150223 08:26:55 <eclesia> good morning
20150223 11:42:12 * doev (~doev@anon) has joined #jogamp
20150223 11:52:28 <zubzub> how can one unit test a opengl application
20150223 11:52:49 <zubzub> in my case I have bytebuffers that represent pixels that are drawn on a rectangle using textures
20150223 11:53:14 <zubzub> (using shaders)
20150223 11:54:02 * zzuegg (~zzuegg@anon) Quit (Quit: Nettalk6 - www.ntalk.de)
20150223 11:56:09 <rmk0> zubzub: seem to remember sgothel mentioning a system where you compare the pixel values of the rendered and saved image and calculate a percentage difference
20150223 11:58:10 <zubzub> sounds nice
20150223 12:00:38 <zubzub> where do I find it? :p
20150223 12:02:37 <rmk0> think it only exists as part of the test suite of some other project
20150223 12:03:04 <rmk0> i have unit tests, but i have a suite of "visual test cases" that i inspect manually every time i make an intrusive change
20150223 12:03:24 <rmk0> http://waste.io7m.com/2015/02/23/manual.png
20150223 12:03:31 <rmk0> it's not ideal, but it's not too awful either
20150223 12:06:04 * zzuegg (~zzuegg___@anon) has joined #jogamp
20150223 12:10:47 <rmk0> (not ideal because it's not automated)
20150223 12:19:59 * jvanek (jvanek@anon) has joined #jogamp
20150223 12:33:27 <zubzub> ideally you'd compare the 2 byte arrays with a fault tolerance
20150223 12:33:38 <zubzub> or something
20150223 12:34:00 <zubzub> or you use a pure software gl implementation who's behavior is know to be stable and predictable
20150223 12:34:13 <zubzub> that way you can at least do some sanity testing
20150223 12:34:47 <rmk0> think that's the main problem... i've no idea how much different implementation vary on a per-pixel basis
20150223 12:42:15 <zubzub> there exists image comparator algoritms
20150223 12:42:49 <zubzub> so you could probably use those and whatever they output to tweak exceptable acceptence criterea
20150223 12:43:20 <zubzub> it'll never be perfect
20150223 12:43:56 <zubzub> but at least you'll be able to spot things like this: http://cdn.arstechnica.net/wp-content/uploads/2014/11/acu-640x400.png
20150223 12:44:46 <rmk0> hehe
20150223 12:55:31 * doev (~doev@anon) Quit (Quit: Verlassend)
20150223 13:02:27 <eclesia> so far in my test cases I test exact pixel values, same result on my nvidia gtx460 and on the amd fireGL
20150223 13:02:45 <eclesia> but those are really simple tests ^^
20150223 13:18:35 <rmk0> thinking about it... i'd sort of hope that different implementations would be near pixel identical
20150223 13:19:17 <rmk0> at least within the core profile without extensions
20150223 13:19:26 <eclesia> the color convertion from vec4 tp bytes should already take care of any minor variations
20150223 16:52:57 * monsieur_max (~maxime@anon) Quit (Quit: Leaving.)
20150223 17:03:03 * eclesia (~husky@anon) has left #jogamp
20150223 17:33:05 * monsieur_max (~maxime@anon) has joined #jogamp
20150223 17:56:28 * jvanek (jvanek@anon) Quit (Quit: Leaving)
20150223 22:01:54 * monsieur_max (~maxime@anon) Quit (Quit: Leaving.)
20150224 05:05:34 -jogamp- Continue @ http://jogamp.org/log/irc/jogamp_20150224050534.html