PTB BETA "goto fail"


“goto fail” is an hommage to Apple’s fine software engineering skills and well known insane attention to detail and perfection.

“goto fail” is the final release of the 3.0.11 series. The next release will start the 3.0.12 series. It contains the following more notable improvements:

  • Our experimental 10 bit framebuffer support for Linux and OSX for AMD Radeon graphics cards is now also available (but untested so far) for Radeon HD-5000 and later graphics cards. In the past this was restricted to X1000 - HD-3000, now it should covers all currently shipping cards since X1000. Along this line goes more fine-grained control of dithering on digital display outputs via new parameters to PsychImaging’s EnableNative10BitFramebuffer functions and a new Screen('ConfigureDisplay', 'Dithering') subfunction; It is important to stress here that while 10 bpc output is known to work reasonably well for analog VGA connected CRT monitors, 10 bpc output to digital display devices will not work on OSX at all, and is unlikely to work over DVI or HDMI on Linux. We are currently looking into ways to enable > 8 bpc support on Linux. What you will get if you enable a 10 bpc framebuffer on such a display setup is a 10 bpc framebuffer which gets reduced to 8 bpc via dithering on the digital video output. Iow., you get simulated 10 bit precision via dithering, which may work reasonably well for some stimuli, or go horribly wrong for other stimuli.

  • Screen('Textfont') now allows atomic setting of font family and and font style with a single Screen('TextFont', window, fontName, fontStyle); call. This to resolve some problems encountered when on tries to use exotic fonts.

  • Eyelink bug fixes and improvements contributed by Li Jie of SR-Research and me. Eyelink('ImageTransfer') should work now, Drawing of targets in PsychEyelinkCallback should work on less capable graphics hardware, Various octave compatibility fixes to SR-Research demos. Eyelink('Command') robustness improvements against some input.

  • Fix Octave + PsychDataPixx long standing compatibility bug for timestamping, remove workaround from and cleanup VBLSyncTest.

  • Workarounds for nouveau-kms bugs in Linux 3.13/3.14 for the upcoming Ubuntu 14.04 LTS release. These bugs shouldn’t have made it into a release but did, because i wasted way too much time fixing Apple’s latest screw-ups, using up most of the testing time i planned to allocate for Linux + Ubuntu 14.04 testing. Anyway, this means that the open-source nouveau graphics driver for NVidia gpu’s will probably be of limited use for vision science until we get the required bug fixes into a post-release update. The proprietary NVidia driver should work just fine though. I will try to not let this happen again. Fixing any problems caused by Apple will have the lowest priority from now on.

  • Workarounds for OSX 10.8 dual-display stereo graphics driver bugs and various 10.8/10.9 graphics bugs. As it turned out during trying to resolve other OSX bugs, at least OSX 10.8 has a hilarious bug when used with two or more open fullscreen onscreen windows, e.g., for dual-display stereoscopic stimulation in stereomode 10. As soon as a 2nd onscreen window opens on a screen, the first previously opened onscreen window will be switched from page flipping for stimulus updated (good for timing and performance) to desktop composition. This is really bad for performance, visual stimulus onset timing and various other things, e.g., 10 bpc framebuffer support. It is also almost completely undetectable by software tests, so will have a good chance of silently corrupted presentation timing in such scenarios. The current PTB will enable a workaround for OSX 10.8 and later which fixed the problem at least on the one tested setup with AMD graphics. It is unclear if this would introduce any new subtle bugs. According to Apple docs it shouldn’t, but of course this is Apple talking about its own products and we know how well that goes in practice. I also realized when testing on my MacBookPro with 10.9.2 and NVidia graphics that 10.9’s swap scheduling on NVidia gpu’s seems to be utterly broken as well. PTB has a new visual test script OSXCompositorIdiocyTest() which you should run on your setups and dual display setups. If that test doesn’t show expected results then you can consider your setup being mostly unuseable for frame accurate visual presentation timing. There are no known workarounds for these issues. I don’t consider “modern” OSX systems trustworthy for frame-accurate visual presentation anymore.

    Bugs not fixed on OSX 10.9: Sometimes you will be left with a blank screen at the end of a session after PTB closed its windows. A mouse click, keyboard press gives you your GUI back. This bug brought to you by Apple. The only known workaround is to use Screen('Preference','ConserveVRAM', 16384); which will likely reduce performance and visual timing accuracy - or possibly make visual timing untrustworthy. I can’t test this because my only remaining Mac test machine has a NVidia graphics card with broken graphics drivers, so timing on my only test machine is always untrustworthy.

    Keyboard queues apparently don’t work at all on recent MacBookAir’s on at least 10.9, because Apple completely broke backwards compatibility in the relevant bits of the OSX HID subsystem.

  • Ability to disable crashy CVDisplayLink timestamping on OSX: PsychTweak('DisableCVDisplayLink'); now allows to disable CoreVideo timestamping as a workaround for timing. If you care about timing you should always install the PsychtoolboxKernelDriver on OSX. CoreVideo timestamping was used as a workaround for setups where this couldn’t be used, e.g., on all Intel graphics cards. The method has proven unreliable due to Apple CoreVideo bugs not fixed since years, but additionally its use causes random crashes at the end of experiment sessions on some setups for some users.

  • PsychImaging - add convenience setup function for dual-display stereo mode 10. See ImagingStereoDemo for the simplified setup for stereomode 10.

  • PsychImaging - Improve 10 bpc framebuffer setup: Do not disable dithering on digital output by default, but leave decision to OS, with option to override.

  • Screen('ConfigureDisplay', 'Dithering') allow low-level control of AMD dither control registers.

  • Various help text updates.

  • Various updates to color calibration/color handling routines by David Brainard.

  • Misc other fixes.

  • PsychtoolboxKernelDriver updates. Different drivers for legacy 10.7 and earlier vs. 10.8+. Apple removed the necessary sdk’s from its latest XCode to build the PsychtoolboxKernelDriver for OSX versions earlier than 10.8. Therefore we now have a for OSX 10.8 and 10.9, which contains bug fixes and improvements. Then a legacy set of 64-Bit and 32-Bit kernel drivers for OSX 10.7 and earlier. These legacy drivers are no longer maintained and contain known bugs and limitation.

  • The OSX Psychtoolbox now requires Octave 3.8.0 or later. It will not work with Octave 3.6 and earlier anymore. It could be made working by manually setting a symbolic link in the file system if somebody wants to stick to 3.6. But upgrading to 3.8 is a good thing, as it now provides a useable GUI and other improvements relevant for PTB.