 |
I have a hunch that this could potentially also be related to 0004341. Just to keep that also in the back of our heads. |
 |
Similarly, when loading, eg a map through the console, the console UI element itself will stretch and flicker for a split frame before going to the load screen. I feel this is likely the same issue as this quicksave HUD bug and my hunch is it has something to do with FBO, but that's a very uneducated guess. |
|
|
So I have run and reloaded tdm with full page heap (enabled via gflags.exe) and I didn't get any memory access violations. While this technique is no guarantee that there are none, access violations are now very unlikely to be the origin of this problem. |
 |
I see this for quite a long time already. I think it is not related to sound problems. This is most likely a purely graphical problem. Some FBO or texture gets rendered after it should not be rendered. I would say that as long as the game does not crash and plays properly afterwards, this issue is about adding "polish" to the game. Unless someone invents a single reliable approach to fix the whole problem, hacking the code to achieve it is not the way to go. |
 |
This wasn't present in 2.06, so it's something we've introduced during this development cycle. Granted, it's not game-breaking, but it certainly looks sloppy if we release an update that introduces such a noticeable glitch. |
 |
During the 2.06 dev cycle, this artifact was commonly seen when r_useMapBufferRange was enabled. In 2.07, there is no cvar to disable it directly but you can sort of change the way it works by disabling fence sync. r_useFenceSync 0 In my testing, this reduced the chance of this showing. Maybe just forcing a glFinish on quicksave would be enough... |
|
|
Binaries are up in SVN. The only time the artifact shows up is if I disabled both uncapped FPS and multi-core and it's very brief. |
|
|
"reduced the chance" sounds exactly like what we do NOT need, i.e. some random incomplete hacks around the code. In the long run such hacks cause pain, and this issue is not severe enough to justify bringing pain upon us. Wouldn't it be better to leave the issue bolder and wait until someone manages to fix it properly, once and for all? |
|
|
He doesn't propose to implement a hack. He just posts his findings. It might be important information that disabling FenceSync affects the issue. :-) Edit: Oh wait, he DID add a hack. That's a nay-nay, NBohr. ;-) |
|
|
In this case, the "hack" is simply a few more glFinish operations. (There is already one in the same code module.) The idea was to polish 2.07 as much as possible, we can revert whatever we want in SVN and make things ugly again. For example, I removed the FBO toggle in the screenshot module (I added to workaround the lightgem breaking on screenshots) ...after the underlying cause was fixed in SVN (after 2.06). It was late in the 2.06 cycle so the idea was "make it work without refactoring large amounts of code". |
|
|
Moved to 2.08 |
|
|
Cannot reproduce in 2.08 SVN latest build |
- Anonymous
