1. 17 Mar, 2019 5 commits
  2. 16 Mar, 2019 7 commits
  3. 15 Mar, 2019 5 commits
  4. 14 Mar, 2019 1 commit
    • Erik Kurzinger's avatar
      wayland/buffer: Try realizing EGLStream before EGLImage buffer · 056c45fe
      Erik Kurzinger authored
      Currently, it is assumed that if querying the EGL_TEXTURE_FORMAT of a
      Wayland buffer succeeds it is an EGLImage. However, this assumption will no
      longer hold on upcoming versions of the NVIDIA EGL Wayland driver which
      will include support for querying this attribute for EGLStream buffers as
      well. Hence, we need to check if buffers are EGLStreams first.
      Fixes #488
  5. 12 Mar, 2019 8 commits
  6. 11 Mar, 2019 6 commits
  7. 08 Mar, 2019 1 commit
    • Carlos Garnacho's avatar
      cogl: Revert swizzling for BGRA buffers · a871d56f
      Carlos Garnacho authored
      As it was originally reported on
      https://bugzilla.gnome.org/show_bug.cgi?id=779234#c0, the hottest path was
      convert_ubyte() in mesa. Reverting this shows no trace of those hot paths,
      nor any higher than usual CPU activity.
      As the improvements at the time were real, I can only conclude that pixel
      conversion was happening somewhere further the pipeline, and swizzling just
      helped indirectly. That got eventually fixed, so swizzling just stayed to
      cause grief. And lots it caused.
      Time to bin this, it seems.
  8. 07 Mar, 2019 1 commit
  9. 06 Mar, 2019 1 commit
  10. 05 Mar, 2019 2 commits
    • Andrea Azzarone's avatar
      sound-player: Don't deadlock in finish_cb · 8856a396
      Andrea Azzarone authored
      The function finish_cb can be called as a result of a call to ca_context_cancel
      in cancelled_cb. This will result in a deadlock because, as per documentation,
      g_cancellable_disconnect cannot be called inside the cancellable handler.
      It is possible to detect if the call to finish_cb is caused by ca_context_cancel
      checking if error_code == CA_ERROR_CANCELED. To avoid the deadlock we should
      call g_signal_handler_disconnect instead g_cancellable_disconnect if this is the
    • Florian Müllner's avatar
      Bump version to 3.31.92 · b0fb39bb
      Florian Müllner authored
      Update NEWS.
  11. 04 Mar, 2019 3 commits
    • Florian Müllner's avatar
      x11-display: Split out restoring of active workspace · 7abceb43
      Florian Müllner authored
      Splitting out the X11 display initialization from display_open() broke
      restoring the previously active workspace in two ways:
       - when dynamic workspaces are used, the old workspaces haven't
         been restored yet, so we stay on the first workspace
       - when static workspaces are used, the code tries to access
         the compositor that hasn't been initialized yet, resulting
         in a segfault
      Fix both those issues by splitting out restoring of the active workspace.
    • Carlos Garnacho's avatar
      wayland: Minor refactor · 3e472faf
      Carlos Garnacho authored
      We use the input_method on both branches, but only check for its existence
      when enabling the text_input. The case of focusing out shouldn't happen in
      practice as we couldn't have focused in ever before, but still make the
      check one level above so it's clearer that the text_input's IM focus cannot
      be enabled without an IM implementation.
    • Carlos Garnacho's avatar
      wayland: Only enable OSK if receiving .enable when already active · eabb7893
      Carlos Garnacho authored
      Enable the OSK if receiving .enable consecutively (i.e. the
      ClutterInputFocus was already focused). We specifically want to avoid
      enabling the panel just because of focus changes within a surface (where
      the .disable request across focus change would previously unfocus the
      ClutterInputFocus). Prior state should be preserved if possible in that