On Wed, Oct 28, 2015 at 10:55 AM, martin rudalics <rudalics@gmx.at> wrote:

> I think I have solved this one. Input events started coming in before the
> init code had finished. By surrounding it by a block_input() and
> unblock_input() pair it seems to work better. I will run it for a couple of
> days and, if the problems have gone away, I will commit it.

Would that help?

Yes, it helped, but for a totally different problem, the startup crash.

I earlier mentioned that here

"on GNUStep making the
toolbar visible in update_frame_tool_bar immediately provokes a call of
windowDidResize which calls updateFrameSize with a not yet updated
toolbar height some time before updateFrameSize gets properly called at
the end of update_frame_tool_bar."

Now update_frame_tool_bar has

block_input ();
...
if (![toolbar isVisible])
[toolbar setVisible: YES];
...
unblock_input ();

and the windowDidResize gets through nevertheless.

I have fixed this by introducing a flag, in_animation, which I set temporarily around calls to setVisible. It makes updateFrameSize return immediately. Problem solved.

BTW, can you reproduce the scenario of bug#21770?

Yes.

I will take a look at fringe-related problems next. I might be able to take a look at this after that.

I just pushed the toolbar fix and the startup crash fix.

-- Anders