fix crash on startup

[zisoft]

In dt_iop_gui_cleanup_module() the mutex gui_lock is destroyed. The orientation module (src/iop/flip.c) has gui_data initialized with NULL, so the cleanup crashes here.

[dterrahe]

Alternatively, make sure the gui->lock is always initialised by moving

dt_pthread_mutex_init(&module->gui_lock, NULL);

from _iop_gui_alloc to dt_iop_gui_init.

That avoids depending on a coincidental relationship between gui_data and gui_lock.

[TurboGit]

Thanks!

[zisoft]

@dterrahe: This does not clearly fix the problem.

Now I get an assertion failure in src/iop/levels.c on startup.

darktable/src/iop/levels.c

Lines 613 to 623 in 526cd08

	void gui_init(dt_iop_module_t *self)
	{
	dt_iop_levels_gui_data_t *g = IOP_GUI_ALLOC(levels);
	dt_iop_gui_enter_critical_section(self);
	g->auto_levels[0] = DT_LEVELS_UNINIT;
	g->auto_levels[1] = DT_LEVELS_UNINIT;
	g->auto_levels[2] = DT_LEVELS_UNINIT;
	g->hash = 0;
	dt_iop_gui_leave_critical_section(self);

dt_iop_gui_enter_critical_section() tries to lock the mutex gui_lock which is now not initialized.

The initialization sequence checks for hidden modules, dt_iop_gui_init() is not called in all cases.

[zisoft]

Revert of ca37ed4 and using dd36220 works.

[TurboGit]

Revert of ca37ed4 and using dd36220 works.

Ok, let's do that for now then. TIA.

[dterrahe]

Fully understood that you don't want to spend time now to find the correct solution to the issue. Unfortunately, my setup seems to be more forgiving and does not throw an assertion on failed initialisation, so I'm not able to debug myself. What's your platform/compiler/profile; maybe I can replicate.

A few thoughts.

• I did not pay too close attention to ca37ed4 yesterday (and assumed it wouldn't make a difference) but the mutex could be initialised before calling gui_init. But also...
• during gui_init I believe no other threads can be accessing "shared" variables yet, so I wonder if dt_iop_gui_enter/leave_critical_section is really required here.
• module->gui_init should not really be called directly, but always via dt_iop_gui_init. As far as I can tell, there's currently only one exception, in _create_deleted_modules which should be converted

      {
        ++darktable.gui->reset;
        module->gui_init(module);
        --darktable.gui->reset;
      }

replaced by

        dt_iop_gui_init(module);

Was that the call that caused your crash?