Itches, Glitches & Anything Else...

[sparhawk]

LOL! Well, it's expensive, but there is a demoversion available, and some *cough* helper tools.

[greebo]

You should really do something about your throat, you're coughing is sounding terrible.

[Fidcal]

I don't mind going through every one in the list again trying to make it happen if it helps. But remind me again about output. I know there is radiant.log in docs and settings. But wasn't there some other output? Or is any of this likely to be helpful?

[greebo]

There is also the console. If DarkRadiant crashes, the console is probably vanishing too fast along with the main app, but you can start DarkRadiant from the command line as well, this way all the output to the console stays after crashing.

 

Don't know if the console will be of much help here, though. It only helps for some assertions ("the application has requested to be terminated in an unusual way").

[OrbWeaver]

The console output could be useful if there is a shared_ptr exception ("Assertion failed: _px == NULL").

[Fidcal]

No - it crashed again but the output just referenced the last model which in this case was the water pump. I suspect you can always get it to fail if you laboriously work through - but whether that is just time. I'll leave the model selector open while I have lunch......

[OrbWeaver]

What exactly did the output say? I am not aware of any standard printouts regarding the selected model, so this could be an error.

 

It is highly unlikely to be a question of clock time, it is more likely to be the number of models you click on, perhaps creating a memory leak or something.

[sparhawk]

Some things you might try:

 

1.

Start DR and click through one model after the other without doing anything else. Do this sufficiently long to see if it crashes after some time. It would also be helpfull to count the clicks for future reference. If it crashes, you should start DR again and do the same, to see if it crashes after the same number of clicks. If that's the case a memory leak is likely.

 

2. Start DR and click again through the models. If you did the previous szenario and you know that it crashes after N clicks, you should stop before you reach that number and do something else. Create some brushes, etc., then continue clicking through the models and see if it crashes again at the same number or at a different time.

 

3. Start DR, do something different (create some entities, brushes, etc.) and THEN start to click through the models. If you did the previous szenarios, check if it crashes at the same count or if there happens something differently.

 

You can repeat this with also browsing through the textures in some variations and see if a similar effect happens here, or if it is only limited to the model browsing.

[Fidcal]

These are the last few lines of the console and in radiant.log...

RenderablePicoSurface: using shader textures/common/collision

RenderablePicoSurface: using shader cutlery_ornamental

RenderablePicoSurface: using shader textures/common/shadow2

RenderablePicoSurface: using shader textures/common/collision

RenderablePicoSurface: using shader tdm_cart01

 

I've done this maybe three times and I'd guess each time it took 3 or 4 or 5 minutes of selecting - the last time with perhaps an hour's lunch in the middle so not just time but as suggested, probably number of selections. I've also mixed mouse clicks with scrolling up and down with the keys - sometimes pausing on a selection; sometimes whizzing past many - so I don't know if they count as selections.

 

Anyway, I'll now try sparhawk's ideas and report later.

[Fidcal]

Made one pass but I'll resume later it is much slower when keeping count and keeping focused so can't face another pass without a break!

 

So far 585 of which 50 are opening + subfolders. All mouse clicks. I doubt I was more than one or two inaccurate so if the next pass approximates that then I think we can accept it's the number of selections.

[greebo]

Wow, you're putting a lot of effort in that. So you get the crash after roughly 600 mouseclicks (550 selections) when using the ModelSelector?

 

I'll have to try that sometime if I can reproduce that. Hopefully this does crash in Linux as well, as in Windows I haven't got a functional debugger.

 

Just a random thought: Could it be that OpenGL is running out of displaylists?

[sparhawk]

Ugh! I didn't think that it would take so long, but that's very dedicated of you. I guess if a developer wants to reproduce the scenario, maybe he can write a loop that simulates what the mouse click does.

[sparhawk]

Wow, you're putting a lot of effort in that. So you get the crash after roughly 600 mouseclicks (550 selections) when using the ModelSelector?

 

Actually it would be much more then 600 mouseclicks. 550 selections would already send 1100 mouseclick events, not counting the clicks to expand the treebranches.

 

But without seeing the code, my estimate is that the up events don't matter much for this problem. Do you cache the models after loading? If not writing a loop to simulate clicking on the same model 600 times should be simple to see if that also makes it crash.

[OrbWeaver]

Just a random thought: Could it be that OpenGL is running out of displaylists?

 

I thought the DL was released automatically when the model is destructed, it is possible I could have forgotten to implement that however.

[greebo]

From looking at the code, the model gets safely released again (and the displaylist in the RenderablePicoSurface destructor), so it's unlikely that this is the cause.

[sparhawk]

Do you have a windows build available built as debug? Or what would I need to setup to compile DR on my system?

[Fidcal]

I've made a second pass and it crashed within two places of last time so I'm assuming that I made some clicking error and its really the same. This time I did not count but just opened everything and clicked everything once all the way down to where it crashed before and I was two short of the place at bc_chest3_lid in containers whereas the first time I just reached the wearables section. (note this is not the same as when I did this before where I was almost randomly moving about.) But I may well have missed one or clicked something twice somewhere.

 

So there is no doubt, the above were clicks including clicking on folders once and then clicking on the + sign on the left - that's two clicks that strictly speaking doesn't select anything (nothing new appears in the 3D view I mean) but I think there are other sub sections like armchair1 that are selections but also have a + sign. I reckon roughly 570 actual model selections.

 

There's a quicker way to crash this iin probably less than two minutes. Open every section in the dark mod models down to darkmod > props > wearables without actually selecting anything then select dark mod at the top and tap the cursor down key quickly. If you just hold the key down I don't think it works as it then it skims over many selections. Just quick taps and it will 'soon' crash.

 

I should mention I am using DR v.0.9.0 Build date April 26 2007 17:45:25.

[greebo]

Do you have a windows build available built as debug? Or what would I need to setup to compile DR on my system?

I haven't built a debug build for quite some time, because it's so much slower. There are some instructions on the wiki about setting up a compile environment, but I don't know if you want to go through that.

 

I've made a second pass and it crashed within two places of last time so I'm assuming that I made some clicking error and its really the same. This time I did not count but just opened everything and clicked everything once all the way down to where it crashed before and I was two short of the place at bc_chest3_lid in containers whereas the first time I just reached the wearables section. (note this is not the same as when I did this before where I was almost randomly moving about.) But I may well have missed one or clicked something twice somewhere.

 

So there is no doubt, the above were clicks including clicking on folders once and then clicking on the + sign on the left - that's two clicks that strictly speaking doesn't select anything (nothing new appears in the 3D view I mean) but I think there are other sub sections like armchair1 that are selections but also have a + sign. I reckon roughly 570 actual model selections.

 

There's a quicker way to crash this iin probably less than two minutes. Open every section in the dark mod models down to darkmod > props > wearables without actually selecting anything then select dark mod at the top and tap the cursor down key quickly. If you just hold the key down I don't think it works as it then it skims over many selections. Just quick taps and it will 'soon' crash.

 

I should mention I am using DR v.0.9.0 Build date April 26 2007 17:45:25.

Thanks Fidcal for your efforts, I will try to reproduce that. Please open a tracker entry for this, so I won't forget about it.

[sparhawk]

I haven't built a debug build for quite some time, because it's so much slower. There are some instructions on the wiki about setting up a compile environment, but I don't know if you want to go through that.

 

Probably not. But I have Purify and SoftICE somewhere, so I might help with it, if you can't reproduce or find it.

[Fidcal]

Thanks Fidcal for your efforts, I will try to reproduce that. Please open a tracker entry for this, so I won't forget about it.

OK - done!

[greebo]

Thanks you two, then.

[jdude]

@jdude: Which type of paste operation did you use? Ctrl-MMB or Shift-MMB? They are doing technically completely different things, that's why I have to know.

 

@jdude & Fidcal: Things happening randomly is not good - I hardly stand a chance of finding that problem. I hope that it happens to me one time so that I can track it down, otherwise I can't really do anything about such problems. It could be anything.

 

The ModelSelector can probably be changed to reload dynamically, I think the VFSTreePopulator can handle that.

 

Could you add these three issues to the bugtracker please?

 

I'm using copy shader and paste shader I have them binded to I and O

[jdude]

The texture tool is great, but it can still be hard to use sometimes. Do you think that it would be possible to have the top of the texture box red, the bottom blue, and the left and right sides different colors? That would make it much MUCH easier to find out which way you have to rotate textures to get them to match up correctly.

[greebo]

What do you mean with bottom, left and right? Do you mean the world orientation of the brush edges? I guess I could get it to work with 4-sided faces, but it wouldn't make sense for polygons that have more than 4 edges. Left and right are non-unique descriptions for a 3D space, however. Or were you referring to something different?

 

Do you have some screenshot of such a situation, so that I can imagine what you're meaning - maybe I can think of another solution.

[greebo]

I just kicked the SmartPointer and SmartReference stuff from the codebase - one weird thing less. This was taking a strange approach anyway, because it delegated the reference counting to the allocated object itself.

 

I also refactored the Doom3Entity class so it's using standard containers now.