Freshplayer often crashes or doesn't work about freshplayerplugin

[swordfeng]

Same behaviour using nvidia gtx755 with proprietary driver

[swordfeng]

console output:
NOT SANDBOXED
NOT SANDBOXED
[fresh 3402] not implemented: PPB_OpenGLES2DrawBuffers(Dev);1.0
[fresh] [error] ppb_message_loop_attach_to_current_thread, thread already have message loop attached
[fresh] [error] ppb_message_loop_attach_to_current_thread, thread already have message loop attached
[fresh] [error] ppb_message_loop_post_quit, bad resource
[fresh] [error] ppb_message_loop_post_quit, bad resource
[fresh] [error] ppb_message_loop_attach_to_current_thread, thread already have message loop attached
[fresh] [error] ppb_message_loop_attach_to_current_thread, thread already have message loop attached
[fresh] [error] ppb_message_loop_post_quit, bad resource
[fresh] [error] ppb_message_loop_post_quit, bad resource

###!!! [Parent][MessageChannel::Call] Error: Channel error: cannot send/recv

[i-rinat]

Yep, I know. But thank you for reporting anyway.
There is ongoing transition to "separated thread" model. Now I have plugin main thread separated from browser thread, which should solve some inherent issues. But since now we have at least two threads, there are synchronization issues and interlocks. And that part is not so great.

There are some thoughts on how to solve these interlocks. And I think I reinvented coroutines. :-)

[i-rinat]

Related issues: #47, #46, #45, #44, #33

[i-rinat]

Please try again with recent version from master branch.

[swordfeng]

Better now.However, it still crashes on a few sites like http://tv.sohu.com/20140803/n403057638.shtml
Chinese video site.There may be many advertisements^_^
log from console:
NOT SANDBOXED
[fresh 5427] not implemented: PPB_OpenGLES2DrawBuffers(Dev);1.0
[fresh] [error] handle_tcp_connect_stage2, evdns returned code 66, count = 0
[fresh] [error] handle_tcp_connect_stage2, evdns returned code 66, count = 0
[fresh] [error] ppb_tcp_socket_private_get_local_address, not connected
[fresh] [error] ppb_tcp_socket_private_get_local_address, not connected
[fresh] [error] ppb_tcp_socket_private_get_remote_address, not connected
[fresh] [error] ppb_tcp_socket_private_get_remote_address, not connected

[i-rinat]

I've reproduced crash on sohu.com. Thanks for the test case.

[i-rinat]

Heh, I'm still not able to fix crashing on tv.sohu.com. There is something wrong happens deep inside libpepflashplayer.so. Most likely it just echoing some heap or stack damage introduced earlier by my code.

So I'm extracting this into separate issue (#60), while closing this one. (It's too tedious to scroll down to the link each time.)

[mittorn]

Freshplayerplugin works on arm only in browser process. Maybe it's better to use valgrind and find memory errors?
I cannot use valgrind on arm because of bad arm instructions support.
I will try to run it on x86 later.

[i-rinat]

I use valgrind on a regular basis. Some recent changes was made just to ease memory corruption detection for valgrind. However all obvious (for me) bugs was already solved that way. Now valgrind shows a bunch of uninitialized memory uses inside libpepflashplayer.so, and then a crash happens. So it does on tv.sohu.com, for example. Just reading wrong location from deep places of libpepflashplayer.so. (With extremely high probability PepperFlash's code is fine though.)

Probably a fresh look at the issue will shed some light on it. I hope so.