IRC Log Viewer » #firebreath » 2012-02-18

IRC Nick Time (GMT-7) Message
EvilTengil 05:02 HIIIII!!
Anyone who have experience getting OpenGL to run (crash free!)?
I posted in the forums
But no response so fart
dougma 05:02 what do you mean that NPP_SetWindow is ignored?
it calls plugin->SetWindow
it being NpapiPluginModule::NPP_SetWindow
dougma 06:02 taxilian: running switcher.html under Chrome i don't see a minor leak (10s of kb), but i have had this assertion firing
the minor leak is constant, the assertion is like 1 in 10 iterations
actually, twice in 70 iterations!
and it seems more easily induced if i switch to another tab
(also, i mean i don't see a major leak, just a minor one)
aha. i am guessing that assertion is due to FBTestPluginAPI::SetTimeout creating a strong pointer to itself.
hmm... commented out the SetTimeout test but still asserting
also SimpleStreamHelper holds a strong pointer to the browser host, that might be it.
trying again with postURL, getURL tests disabled.
50 iterations no probs, that'll be it. i'll make it a weak pointer
dougma 06:02 fixed. :)
removed it altogether
EvilTengil 06:02 dougma, it's just calling plugin->SetWindow the first time. But the other times, when the client rect has changed but the HWND is the same, it doesn't send any signal to the plugin.
And it seems like that when the plugin clients visible rect has changed, the DC is invalidated and the following ogl calls crashes the plugin!
dougma 06:02 client rect changing is a different codepath
EvilTengil 06:02 But that event seems to sent to the plugin by NpApi by calling NPP_SetWindow, no?
dougma 06:02 oh ok i was thinking activex
EvilTengil 06:02 in activex it's working :)
its's in chrome it crashes
dougma 06:02 cause activex is good (compared to npapi)
EvilTengil 06:02 and i put a breakpoint in the NPP_SetWindow handler in FireBreath and i get that event exactly before everything crashes, and the following ogl calls crashes the plugin!
As soon as any part of the plugin is scrolled outside the viewbow/scrollbox or whatever it's called of the browser that event is sent and the ogl rendering crashes!
dougma 06:02 i just wouldn't trust using a DC cross-thread
EvilTengil 06:02 hmm im using it only in one thread
dougma 06:02 oh yes... i reread your forum post
it's late, i've had a beer. :)
EvilTengil 06:02 but its the ogl thread
could it be a problem that im actually grabbing the DC from another thread than the thread that is sending the Attached event?
How threadsafe is win32..?
dougma 07:02 the threa where the DC is created is where you should use it
wellll... that's not quite what it says here: Note that the handle to the DC can only be used by a single thread at any one time
i mean here:
so yeah if you're using it on your own opengl thread then i would think you're asking for problems.
but i've only ever done windowless-drawing and never opengl, so...
also remember that chrome is going to extraordinary lengths to run your plugin out-of-process
IE's isolation model is different.
have you had any success with Firefox? firefox is interesting to test these things cause you can turn ipc on and off.
EvilTengil 07:02 i havent
i guess i should
dougma 07:02 still npapi of course
EvilTengil 07:02 its working fine in firerfox!
only chrome then.. gag
dougma 07:02 ipc on (the default) or off ?
EvilTengil 07:02 ipc?
i havent changed any settings
dougma 07:02 probably the default then. :)
that's good news. boo chrome!
EvilTengil 07:02 it seems really hard to write browser plugins that just work. different browsers, versions, operating systems.. wondering if it's really worth it
dougma 07:02 it's a nightmare but i get paid to do it. :)
EvilTengil 07:02 me too but with all this crap i should get paid more haha :P
dougma 07:02 i always advise the business no to do it.
but they insist
EvilTengil 07:02 my project is more of a research project. the reason why we thought it would be good with a browser plugin is that the users should feel it was a bit less intrusive and lighter than installing a standalone application
and instead of having a ugly Qt interface or something it could all be done in the browser (except for the actual video feed im trying to render with ogl)
but as it looks right now i think im gonna implement unaccelerated GDI blitting instead
it should be fast enought to display a 10-30 fps 640x480 webcam feed right..?
dougma 07:02 yeah... that's not without issues too, but it is pretty fast
yeah it'll do that
EvilTengil 07:02 ok good
dougma 07:02 have you done windowless plugins before?
EvilTengil 07:02 nope
this is my first plugin ever
it was supposed to be done by this other guy who then bailed leaving me with it
dougma 07:02 ok. yeah so you call 'invalidate' and the browser calls you back to draw.
EvilTengil 07:02 i did a small test
but i realized that was not enough
dougma 07:02 ie will tend to call multiple times to repaint the window in a striped fashion
EvilTengil 07:02 because the erase background event it flickers so i think i need to implement my own event handler and discard those erase events
ah windowless
this was just for windowed GDI
dougma 07:02 yeah, windowless doesn't flicker and interacts with other page elements better
EvilTengil 07:02 seems slow? can you invalidate just the region so that not the entire browser window is being repainted=
dougma 07:02 it just repaints your plugin
you call the browser-plugin interface to invalidate
EvilTengil 07:02 is it supported in all browsers?
dougma 07:02 yup. and the only option on mac
EvilTengil 07:02 oh interesting
i dont really have much gui
i actually just want to blit the images im getting from this kinect library i wrote. so it's just blitting and perhaps writing some text like "Please wait..." and error messages that has to be drawn. would probably be just as easy to just implement native drawing for every platform than using ogl.
dougma 07:02 yeah
EvilTengil 07:02 ogl would just be nice because then it would be easy to support drawing the pointcloud live which would be cool
dougma 07:02 and use html for the messages
EvilTengil 07:02 but that's a feature that would be saved for later nayway
oh thats right... you can draw stuff above windowless things
that would make it much easier to make a nice loading animation :))
dougma 08:02 taxilian: that leakfinder in fbtestplugin is cool! first time i've used it.
and yeah, leaking like a sieve!
oh maybe.... not the first one is a static. will go through one by one. :/
dougma 08:02 leakfinder is just picking up all the static instances... i can't see any leaks (running fbtestplugin)
taxilian 08:02 dougma: it is possible the problem was in firefox, but I ran FBTestPlugin with the switcher page for about 2 hours and the process was up to 4 gig
dougma 08:02 oh, will try ff.
taxilian 08:02 I did it again w/out your fix for comparison (quite possible it was already there) and just haven't looked at the results
it's probably down there still running :-P
I put an offer on a house yesterday, so I've been a bit distracted
dougma 08:02 no probs
i'm about to crash
maybe i should do release build too
ff (8.0.1) plugin-container uses way more ram than chrome's!
but not growing
taxilian 08:02 hmm
I'll keep playing
dougma 08:02 the mem leaker keeps growing
visual studio is too slow, i'm off!
taxilian 09:02 hehe. g'night =]
dougma 09:02 oh yeah, i am running with that simple-stream change too...
really off now.
FireBreathBot 11:02 JIRA issue issue commented by thesheeep "I have the same error and use Cmake 2.8.3. That shouldn't be too old."
FireBreathBot 13:02 JIRA issue issue commented by richard "Just try upgrading and verify, please; there was a version of cmake that I had this problem with,..."
FireBreathBot 14:02 JIRA issue issue commented by thesheeep "Okay, I tried with 2.8.7 and the error does indeed not appear there.
JIRA issue issue comment edited by thesheeep "Okay, I tried with 2.8.7 and the error does indeed not appear there.
FireBreathBot 14:02 JIRA issue issue resolved by richard "It seems that the fix is to upgrade to the latest cmake; if that doesn't work, let me know. This ..."
ch 17:02 taxilian: might be worth bumping cmake min version to 2.8.7 then?
taxilian 17:02 the problem only happens on windows
ch 17:02 hm, okay