Application and system freeze on Windows

[rawnsley]

After rendering any model (F5 or F6) any attempt to rotate or translate the viewport (using the mouse) causes the application to freeze. It also causes Windows to become completely unresponsive to everything but CTRL + ALT + DELETE, which curiously brings OpenSCAD back to life and allows it to move the viewport. However, the problem occurs again as soon as the viewport is moved. Sometimes only Task Manager can save you from having to do a hard reset.

I'm using Windows 7 Home Premium 64 Bit Service Pack 1. I understand (from my Google searches) that other Windows users have experienced similar problems.

[donbright]

i cannot reproduce this bug on WinXP. i do not have access to any windows 7 machine. if someone wants to provide remote access to a windows 7 machine i will be happy to look into this bug.

[rawnsley]

Windows XP! The fat-cats at OpenSCAD clearly aren't paying you enough ;-)

The nature of the problem is such that it would probably lockup any remote desktop session. We could try a Google+ Hangout via my phone and I can point it at the screen. This is me: https://plus.google.com/118363380530058848433

[donbright]

Sorry but I will need 50+ hours of interactive remote desktop access to an actual machine to do anything useful. My first step would be to try to compile QT and OpenSCAD using an updated version of a free MSVC:

http://en.wikibooks.org/wiki/OpenSCAD_User_Manual/Building_on_Windows
https://www.google.com/search?rls=en&q=qt+crash+windows+7

I would also test the cross-compiled MINGW/GNU version of OpenSCAD vs the MSVC compiled version to see if it happened on both.

I would probably also try compiling with vs without optimization flags to the compiler.

IIRC the remote desktop programs include 'send ctrl-alt-delete' function in their window menu.

[rawnsley]

Thanks for trying to repro anyway. Not sure we can do much without a budget for a real machine or rental time on virtual Windows machine. It still smells like a problem that won't manifest remotely, but I will experiment at home and add anything I find that might help.

[donbright]

Ok some more ideas.

1. Does this bug happen running the 32-bit version of OpenSCAD on Windows? Or only 64 bit?
2. Im downloading Windows 8 free developer preview 32-bit + visual studio express 32-bit. If all this stuff installs on my computer, maybe i can see the bug
3. Im cross-compiling with various settings. Here is one with standard mingw-cross-env, but openscad compiled without optimization:

https://github.com/downloads/donbright/openscad/openscad.zip

If someone could try it it would be cool.

[donbright]

hit a bit of a problem installing Windows 8 developer preview. when you go to install the Visual Studio Express, here is the license you have to agree to:

"If you comply with these license terms, you have the rights below.

1. INSTALLATION AND USE RIGHTS.
   •
   You may install and use any number of copies of the software on your devices solely for evaluation purposes to design, develop and test your programs for use with the software.

•
You may not deploy or distribute any program you design or develop with the software, except that you may deploy your programs internally solely to evaluate any deployment technologies contained in the software.

"

So if i am using it to fix a bug that is probably 'development' and if i upload the patch to github, thats probably 'distributing'. So that plan will not work. Back to cross-compiling.

[rawnsley]

Frustratingly it has stopped happening after some OS updates and a reboot. Here are the updates: http://imgur.com/Xcnre, but it might simply have been the reboot.

In answer to your questions: I've only seen the problem on my ACER laptop running 64 bit Windows. I tried the other version you uploaded on 31 December, but it also locked up (and is now working OK).

I will make some designs in OpenSCAD and see if it happens again. If I can't reproduce it within a reasonable time I will close the issue.

[donbright]

thanks

i just tried the 32 bit release version of openscad for windows on Windows 8 developer preview 32-bit and it works, although it is a little choppy (not smooth).

[kintel]

Just to clarify what is going on, could you do the following?

1. Help->OpenGL info: Copy and paste the information you see and attach it to this issue
2. Does this happen with every model you try to render? Even a simple cube() ?
3. Are you sure it happens both with F5 and F6? Can you turn off "Automatic Reload and Compile" and render using F6 only to verify?
4. Can you try this on a freshly booted machine without running other applications?

[rawnsley]

It's actually stopped happening now, possibly because of a Windows Update (see previous post), but there was nothing obvious in the list of update packages, expect perhaps the VC Runtime security update: maybe my VC Runtime files were corrupt and this update patched them? If it starts happening again I will be straight back on here with more info.

1. I've added the OpenGL dump to the bottom of this post
2. It happened even with very simple shapes. The simplest manifestation was a long flat cube with a cylinder stuck to the top of it.
3. It definitely happened with both F5 and F6, which lead me to believe it was a problem with the graphical sub-system as opposed to the OpenSCAD model processing. I never run with Automatic Reload and Compile, but I'm not sure I ever did an F6 without having previously done an F5. I will experiment in future if the problem comes back.
4. It certainly has happened with no other programs running, but the only machine I saw it on is my development laptop, and that has a large number of installed programs and platforms, any of which might be running background processes or have disturbed the delicate balance of system files.

GLEW version 1.7.0
OpenGL version Intel(R) HD Graphics
Intel (2.1.0 - Build 8.15.10.2559)

RGBA(8888), depth(24), stencil(8)
Extensions:
GL_EXT_blend_minmax GL_EXT_blend_subtract GL_EXT_blend_color GL_EXT_abgr GL_EXT_texture3D GL_EXT_clip_volume_hint GL_EXT_compiled_vertex_array GL_SGIS_texture_edge_clamp GL_SGIS_generate_mipmap GL_EXT_draw_range_elements GL_SGIS_texture_lod GL_EXT_rescale_normal GL_EXT_packed_pixels GL_EXT_texture_edge_clamp GL_EXT_separate_specular_color GL_ARB_multitexture GL_EXT_texture_env_combine GL_EXT_bgra GL_EXT_blend_func_separate GL_EXT_secondary_color GL_EXT_fog_coord GL_EXT_texture_env_add GL_ARB_texture_cube_map GL_ARB_transpose_matrix GL_ARB_texture_env_add GL_IBM_texture_mirrored_repeat GL_EXT_multi_draw_arrays GL_NV_blend_square GL_ARB_texture_compression GL_3DFX_texture_compression_FXT1 GL_EXT_texture_filter_anisotropic GL_ARB_texture_border_clamp GL_ARB_point_parameters GL_ARB_texture_env_combine GL_ARB_texture_env_dot3 GL_ARB_texture_env_crossbar GL_EXT_texture_compression_s3tc GL_ARB_shadow GL_ARB_window_pos GL_EXT_shadow_funcs GL_EXT_stencil_wrap GL_ARB_vertex_program GL_EXT_texture_rectangle GL_ARB_fragment_program GL_EXT_stencil_two_side GL_ATI_separate_stencil GL_ARB_vertex_buffer_object GL_EXT_texture_lod_bias GL_ARB_occlusion_query GL_ARB_fragment_shader GL_ARB_shader_objects GL_ARB_shading_language_100 GL_ARB_texture_non_power_of_two GL_ARB_vertex_shader GL_NV_texgen_reflection GL_ARB_point_sprite GL_ARB_fragment_program_shadow GL_EXT_blend_equation_separate GL_ARB_depth_texture GL_ARB_texture_rectangle GL_ARB_draw_buffers GL_ARB_color_buffer_float GL_ARB_half_float_pixel GL_ARB_texture_float GL_ARB_pixel_buffer_object GL_EXT_framebuffer_object GL_ARB_draw_instanced GL_ARB_half_float_vertex GL_EXT_draw_buffers2 GL_WIN_swap_hint GL_EXT_texture_sRGB GL_EXT_packed_float GL_EXT_texture_shared_exponent GL_ARB_texture_rg GL_ARB_texture_compression_rgtc GL_NV_conditional_render GL_EXT_texture_swizzle GL_ARB_sync GL_ARB_framebuffer_sRGB GL_EXT_packed_depth_stencil GL_ARB_depth_buffer_float GL_EXT_transform_feedback GL_EXT_framebuffer_blit GL_ARB_framebuffer_object GL_EXT_texture_array GL_ARB_map_buffer_range GL_EXT_texture_snorm GL_INTEL_performance_queries GL_ARB_copy_buffer GL_ARB_sampler_objects GL_NV_primitive_restart GL_ARB_seamless_cube_map GL_ARB_uniform_buffer_object GL_ARB_depth_clamp GL_ARB_vertex_array_bgra GL_ARB_draw_elements_base_vertex GL_EXT_gpu_program_parameters GL_ARB_compatibility GL_ARB_vertex_array_object

[kintel]

Another thing to test if this bug arises again:
Set the environment variable OPENSCAD_DISABLE_GL20 to 1.
This will turn off all OpenGL 2.0 features, just in case the drivers barf at some FBO or shader code.

[rawnsley]

Its happening again. Adding the OPENSCAD_DISABLE_GL20 variable didn't make a difference; I know it was active because there is a warning dialog on start-up. I'll try a reboot and report back if it helps, otherwise I'm happy to try other things if you have any suggestions.

[donbright]

is there any chance it is your video drivers?

[rawnsley]

I was on the latest version, but I did a reinstall to be sure (in case any DLLs had become corrupt). The problem has gone away, but it might have been the reboot. I will try it periodically and let you know how it goes.

[donbright]

i managed to get an amazon AWS remote desktop of Windows Server 2008, but i still cannot reproduce this problem.

my best guess right now, is that there is some kind of bad interaction between the video drivers and some part of openscad (maybe qt?). if there was some way to switch to 'software rendering' it would narrow it down somewhat, but i am not sure how you do that in windows.

[donbright]

there are several people on the mailing list having some similar issues. here is what Nop Head wrote:

"I just had it happen to me again when I had my Python machine control script running. It always causes the lockup, but the interesting thing is that it still does it during the bed cool down routine, which is only polling the machine very second. In between it is sleeping, so how it manages to have such a dramatic effect on Openscad I have no idea. It doesn't even have a GUI, just Ethernet UDP comms, a few packets every second. As soon as it terminated openscad un-froze."

Here is Pasca Andrei

"Ah, OK...This was the first time for me when it happend - I had an MS Fffice 2007 document oppend. After I closed it, the problem was gone. And that could be a big problem for me as I can't control what applications are running when openscad is launched by others with my models..."

I am beginning to think maybe a good way to track down this bug might be to write some kind of non-QT interface for openscad and see if it dies then too - at least that will narrow things down. But that is probably a few months off.

[eegan]

To remove any doubt about 32 bit versus 64: it happens reliably on my 32 bit Windows XP Professional IBM/Lenovo R60e. On the other hand it never happens on my Dell Dimension 2400 (Windows XP home), or on another more recent Dell running Vista.

I don't think I'm the only one who has said this, but the problem seems consistent with a hijacking of most user input messages. It is produced by clicking in the viewport. Note that I don't even have to have anything rendered to produce the problem. I can open OpenSCAD, click 10 times in the (empty) viewport, and produce the problem seemingly 100% of the time.

Ctrl-alt-del followed by an ESC (cancel) usually unblocks it. This seems consistent with a user input issue, since the ctrl-alt-del message, being the ultimate meta-key, is handled at a different or lower level. Sometimes I have to start the process manager and kill OpenSCAD to unblock the system.

The problem does not seem to stop anything other than user input messages. If I have the Windows clock running and visible, I can see the second hand keeping on ticking while the problem occurs.

When it happens, the Windows key toggles the start menu, but other keys e.g. the cursor keys that normally let you cycle among start menu items, are ignored. So it is affecting some but not all keystrokes as well as mouse clicks, once again pointing to a user input message processing issue.

Whether the problem happens in a remote session is an interesting question, but I think it's quite likely it won't be debuggable in a remote session: if it happens, it probably won't be possible to recover from it unless a ctrl-alt-del can be simulated on the remote machine. I can try this as well.

I ran Spy++ to look at system messages and I can confirm that after the problem occurs, system messages from mouse clicks and movements and key presses stop appearing as posted to the OpenSCAD window, and also to the desktop, as two examples I tried.

[kintel]

Thanks for the thorough report.

I'm starting to think that this is triggered by one of two things:
o Build environment: mingw vs. Visual Studio
o Buggy OpenGL drivers in combination with our GL code

It would be interesting to try the build environment path first.

Don: Did you ever get a Visual Studio build up and running?

[donbright]

no, i dont have access to a windows machine anymore. sorry. i havent been able to do much lately, alot of stuff happening in 'real life'.

[shadowjig]

I have this same issue on a laptop running Windows Vista (32 bit). All drivers (touchpad, display, etc) are up to date. If I switch to my desktop PC it works fine. I believe this is an issue with the touchpads on laptops.

[eegan]

On my Windows XP laptop it happens equally with a USB mouse.

[Karijn]

Hi,

I've exactly the same problem. I tried older versions of OpenScad too with the same results.
I've a Windows 7-64bit HP-Laptop

I do use VS-2010 and some Phyton programs, but not as the same moment

[hedgren]

I had this same issue on my Windows 7 x64 Dell Laptop. When I googled it I found a post that indicated it was related to Microsoft's Intelipoint. So, I uninstalled Intellipoint and rebooted my system. It now seems to work fine for me. I hope this helps.

[xnaron]

Have the developers been able to replicate this issue? I have been suffering with this issue for a year now. Running Windows 7 64bit. Openscad will work fine after a reboot and eventually it will start freezing up. I'll need to ctl-alt-delete and then the desktop will unfreeze. Then I need to reboot again before I can use openscad. I tried uninstalling intellipoint and I still have the problem. This is a really frustrating problem that seems to be effecting a lot of users. I'm really hoping the dev's will be able to recreate the issue and find a solution.

[jeremydok]

_Edit: If you are having this problem please read my comment below and install the Microsoft re-distributable instead of python to help narrow down the problem_

I was having this problem on a Win7 32bit. I tried several different openscad versions with the same result. I kept the OS upto date with no solution. I had to keep using Ctrl-Alt-Del and then the task manager to kill it.

What fixed it properly was installing the latest version of Python on the machine. I saw the thread at
http://forums.reprap.org/read.php?80,94602,94602,quote=1
I installed python version 3.3.0 which works great now.

[donbright]

that is really bizarre - the Openscad GUI does not use python. if python is the problem, then stopping all running python programs should allow OpenSCAD to work regardless of the version of python installed. Unless there is some hidden program running in the background using python?

as nophead pointed out, python3 is not very compatible with python2, it seems like some users might have an issue doing this upgrade since their python2 programs might not work?

[kintel]

Hm, perhaps the python installer updates some system libraries which we use? Sounds dubious though.

[jeremydok]

_Edit: If you are having this problem please read my comment below and install the Microsoft re-distributable instead of python to help narrow down the problem_

Very curious indeed. I was was following the advise given in the link above. I had openscad installed for maybe a month, tried to get it working a dozen different ways with no luck; then tried the python install and it works every time. Could someone else who is getting freezes give it a try and post results. kintel might be on to something as others have reported other updates fixing it as well.

[shadowjig]

I've consistently had issues on my laptop, but my desktop works fine. I installed Python 3.2.3 on my laptop and now I can use the zoom in and out function without having to CTRL-ALT-DEL. I'm running Windows Vista Home 32-bit and OpenSCAD 2012.04.01.

Thank you for the suggestion, installing python 3.2.3 now makes OpenSCAD usable on my laptop.

[xnaron]

Be interesting to see long term if you don't encounter the issue anymore. I uninstalled the intellimouse in the control panel and it is less frequent. Before I needed to reboot when it started happening. Now I can usually just close running apps and I can get it responding again. Still a pain but better than having to reboot.

[AustinSpafford]

And confirmed: Rebooting my system (no updates included) alleviates the symptoms. If they return, I will post again with an approximate log of the events leading up to the recurrence.

[AustinSpafford]

After only running a single instance of OpenSCAD and Chrome for just a few hours, the symptoms have returned Also, task manager shows that one of the CPU cores goes to nearly 100% usage during the period of input-lockout. More specifically, upon using Control+Alt+Delete->Escape to restore control, the histograms show a period of CPU-grind.

Is it possible that there's a tight-loop within the mouse-look code? What sort of hook is being used to enter a cursor-less delta mode? Is it possible that the mouse-look code is performing some sort of calculations based on the OS up-time?

[donbright]

AustinSpafford: most of the mouse code is here:

https://github.com/openscad/openscad/blob/master/src/GLView.h
https://github.com/openscad/openscad/blob/master/src/glview.cc

It basically subclasses QT's QGLWidget and uses some matrix math to change the OpenGL translation matrix.

http://qt-project.org/doc/qt-4.8/QGLWidget.html
http://www.opengl.org/sdk/docs/man/xhtml/glTranslate.xml
http://eigen.tuxfamily.org/dox/classEigen_1_1AngleAxis.html

[MichaelAtOz]

I have this issue too. W7Ultimate/64.

I have 7 MS VC++ Redistributables installed;

2 x 2005(x64) {0.0.56336, 8.0.61000},
5 x 2008 - x64 {0.0.30729.17, 9.0.30729.4128, 9.0.30729.6161, 9.0.30729.4148, 9.0.30729.6161}.

That's interesting, didn't notice the duplication until I had to type it...also the dups have different sizes (??) I'll try uninstalling the dups and confirm whether it still occurs, noting that these were installed as dependencies when I installed other stuff (god knows what). TBC. On second thoughts I won't do this yet as they will be version dependencies of the other products.

My observations are much like Austin's.

I can get it with a newly started OpenSCAD without a file loaded. Hence no Compile or Render (auto compile is off). Basically with OpenSCAD not the foreground task, one click on the yellow pane and pop - up goes the CPU 50% of dual core and input blocked except for the Mouse movement (no display highlighting happens when moving mouse over icons that normally do something on a hover), Windows key & Ctl-Alt-Del. Other tasks are still running eg I have a sidebar-ish thing which is happily updating various info & Task Manager is still updating it's window.

One or sometimes two cycles of Ctl-Alt-Del & Esc gets it back, then it may behave for a while, but sometimes happens every time again for a while.

I have just tried a few things, it appears that if I click on the windows title bar to bring it to foreground before clicking on the yellow pane it doesn't happen. So it may be due to the windows Activation messages WM_ACTIVATEAPP and/or WM_ENABLE not being handled appropriately?? ie I think the owner process of the yellow pane may not be removing a message from the message queue hence all input messages are queued. This is consistent with what happens after C-A-Del, as all the mouse movements (drags etc) in the yellow pane happen, like a replay, just after C-A-Del and before the C-A-Del-Blue-Screen pops up.

I just tried repeatedly activating OpenSCAD by clicking on the console area before the yellow pane, it does not cause the issue.

I had Task Manager open with the OpenSCAD process highlighted, just as it happens, the Working Set has a ~8-12M drop (only with a model loaded), don't know if that is relevant.

I hope that help track something down.

I'm not an in-depth Windows Programmer, but am reasonably proficient, let me know if I can help some more.

Michael

[MichaelAtOz]

ps For others, that may be a work-around.

Click on the title bar or the console to activate the OpenSCAD window.

Seems to be working for me.

[donbright]

Ok. here is a question. Giles Bathgate's RapCAD has many similarities to OpenSCAD, for example it has a QWidget of type GLView that has many of the same functions.

Does RapCAD have the same issues as OpenSCAD on these boxes? Does it freeze up? Does the system lock?

[MichaelAtOz]

Update, all it takes is the Mouse_Down, doesn't need the Mouse_Up to cause it.
Also happens with a left, right or middle button.

I'll give RapCad a go.

[MichaelAtOz]

Well there you go, it happens with RapCad too.

[MichaelAtOz]

In https://github.com/openscad/openscad/blob/master/src/glview.cc

void GLView::mousePressEvent(QMouseEvent *event)
{
mouse_drag_active = true;
last_mouse = event->globalPos();
grabMouse();
setFocus();
}

Just a hunch; don't use grabMouse at all; as per the doco;

**void QGraphicsItem::grabMouse ()

Grabs the mouse input.

This item will receive all mouse events for the scene until any of the following events occurs:

The item becomes invisible
The item is removed from the scene
The item is deleted
The item call ungrabMouse()
Another item calls grabMouse(); the item will regain the mouse grab when the other item calls ungrabMouse().

When an item gains the mouse grab, it receives a QEvent::GrabMouse event. When it loses the mouse grab, it receives a QEvent::UngrabMouse event. These events can be used to detect when your item gains or loses the mouse grab through other means than receiving mouse button events.

It is almost never necessary to explicitly grab the mouse in Qt, as Qt grabs and releases it sensibly. In particular, Qt grabs the mouse when you press a mouse button, and keeps the mouse grabbed until you release the last mouse button. Also, Qt::Popup widgets implicitly call grabMouse() when shown, and ungrabMouse() when hidden.

Note that only visible items can grab mouse input. Calling grabMouse() on an invisible item has no effect.

[MichaelAtOz]

Sorry, don't mean to spam, but I found this

Warning: If you call setFocus() in a function which may itself be called from focusOutEvent() or focusInEvent(), you may experience infinite recursion.

Infinite recursion sounds like what is going on.

[kintel]

I committed a fix which removes grabMouse() and setFocus() as I indeed didn't see the need for those calls.
Could someone create a Windows build and see if this has improved things?

[xnaron]

That will be so awesome if this fixes the issue. It is so annoying.

[donbright]

i am not very experienced with the cross compile but here is a bare bones openscad.exe made with Marius' new patch:

https://github.com/downloads/donbright/openscad/openscad.zip

I am not even sure how many 'bits' it is. but it works on Wine.

[MichaelAtOz]

Quick test looks good so far. Thanks.

The console is a bit more wordy, presumably debugging level (PolySetCache hit:... CGAL Cache insert...)?

Will report back after more extensive use.

ps running as 32bit app.

Seems to run quicker, particularly Rendering.

[eegan]

Looks good indeed.

I tried without success to make the new version freeze up. Just to be sure nothing had changed (in the system) I ran the old .exe and got it to freeze within a few seconds of clicking around in the view window.

It seems pretty likely to me that this fixes the problem.

I am going to go out on a limb and say thank you to everyone who helped diagnose and fix this. I will now get down to learning OpenSCAD now that it is usable on my machine!

[MichaelAtOz]

Well I have not had 'the' issue with the fix, but I had others which I don't with the old version (2011.12.30).

I just had a crash, just after loading and autocompiling a .scad

Problem signature:
Problem Event Name: APPCRASH
Application Name: openscad.exe
Application Version: 0.0.0.0
Application Timestamp: 500230f6
Fault Module Name: OLE32.dll
Fault Module Version: 6.1.7601.17514
Fault Module Timestamp: 4ce7b96f
Exception Code: c0000005
Exception Offset: 0002251d
OS Version: 6.1.7601.2.1.0.256.1
Locale ID: 3081
Additional Information 1: a7aa
Additional Information 2: a7aa91f17ea749d42a4de3b390fa5b3d
Additional Information 3: a7aa

Additional Information 4: a7aa91f17ea749d42a4de3b390fa5b3d

That was after loading working on various files.

Rerunning the fixed OpenSCAD, then reloaded the same file, autocompile.
It did not display the text on the item, which the old version does.

When I did a Render I got:

MS V C++ Runtime Library Dialog Box;
Assertion failed!
Program: ...openscad.exe
File: /home/dond/mxe/usr/lib/gcc/i686-pc.../shared_ptr.hpp
Line: 424

Expression: px != 0

For info...

Abort Retry Ignore

Should I lodge these separately?

<sorry, didn't know the lines would do the loud formatting>

[donbright]

ouch. please do file a bug. if there is any reproducibility, or small example .scad, it would be very helpful.

[MichaelAtOz]

Issue #148 raised for above.

[nophead]

Seems to fix the lock up but it prints a lot of debug information about cache hits.

Also 0.8 seems to now print as 0.80000000000000004 in an echo, which really messes up my BOM generation.

[kintel]

The debug info is because don's build is a debug build and should be gone in the next release build.
I'll open a separate issue on the echo issue.

[nophead]

Some values used to do it before. I think recurring binary fractions don't get rounded nicely. I can't think why it changed unless this version uses a different floating point library.

[kintel]

There was some refactoring of the variants in the backend. That might have caused it.

[kintel]

I believe this has been fixed now - closing.

[donbright]

For anyone viewing this, a working non-freezing Windows installer available at http://www.openscad.org . Please choose a version newer than July 2012 (Either the Development Snapshot or any releases made after July 2012)