config_section=old

no-reply@allegro.cc (Matthew Leverton) — Sun, 22 Jan 2012 09:22:05 +0000

xglx_config.c#SelectExpand
 378staticvoid select_best_visual(ALLEGRO_DISPLAY_XGLX *glx,
 379                               ALLEGRO_EXTRA_DISPLAY_SETTINGS**eds, int eds_count,
 380bool using_fbc) 381{ 382   ALLEGRO_SYSTEM_XGLX *system=(void*)al_get_system_driver(); 383 384qsort(eds, eds_count, sizeof(*eds), _al_display_settings_sorter); 385 386   ASSERT(eds[0]); 
 387 388if(!eds[0]->info){ 389       ALLEGRO_ERROR("No matching displays found.\n"); 390       glx->xvinfo = NULL; 391return; 392}

Using VirtualBox, the assert (added by me) is triggered. All 64 visuals are "incomplete", but it still says eds_count=64.

xgl_config.c#SelectExpand
 226   ALLEGRO_INFO("%i formats.\n", num_fbconfigs); 227 228for(i =0; i < num_fbconfigs; i++){ 229      ALLEGRO_DEBUG("-- \n"); 230      ALLEGRO_DEBUG("Decoding visual no. %i...\n", i); 231      eds[i]= read_fbconfig(system->gfxdisplay, fbconfig[i]); 232if(!eds[i]) 233         continue; 
 234#ifdef DEBUGMODE 235      display_pixel_format(eds[i]); 236#endif 237      eds[i]->score = _al_score_display_settings(eds[i], ref); 238if(eds[i]->score ==-1){ 239continue; 240} 241      eds[i]->index= i; 242      eds[i]->info =al_malloc(sizeof(GLXFBConfig)); 243memcpy(eds[i]->info, &fbconfig[i], sizeof(GLXFBConfig)); 244} 245 246*eds_count = i; 247   ALLEGRO_INFO("%i visuals are good enough.\n", i);

Could somebody who is familiar with this clean that up? It seems very sloppy to ignore NULL values, and still consider eds_count to equal i. It appears to me that you'd get 64 NULLs which are happily passed along.

I could hack in a check to switch to the old function if everything is NULL, but that isn't very proper.

With the following config, it works:

[graphics]
config_selection=old

no-reply@allegro.cc (Peter Wang) — Sun, 22 Jan 2012 10:24:05 +0000

You mean like this? (not really tested)

no-reply@allegro.cc (Matthew Leverton) — Sun, 22 Jan 2012 10:28:29 +0000

Cannot really tell from that patch, but does this mean it will automatically fall back to "old" if "new" fails? I don't even know what's going on in that file or what the difference is between the two, so pardon my lack of effort in trying to do any debugging.

no-reply@allegro.cc (Peter Wang) — Sun, 22 Jan 2012 10:38:02 +0000

It should do (fall back) although that wasn't what I was going for. Actually I wasn't sure what your bug report was about.

no-reply@allegro.cc (Matthew Leverton) — Tue, 24 Jan 2012 05:46:56 +0000

Running Ubuntu under a VM triggers the behavior of those NULL values and crash on every created display.

Using the config to skip that function "fixes" it. People complain about it from time to time (not sure under what environment), and usually Elias stops by to tell them to create an allegro5.cfg with the above settings.

Update:

The patch seems to work. I just applied it as-is, and the examples all run.

no-reply@allegro.cc (Peter Wang) — Thu, 26 Jan 2012 04:49:25 +0000

I always assumed that the crash was due to a bug in the newer GLX functions, otherwise why did Milan make it a configurable option? I will apply the patch.

no-reply@allegro.cc (Don Freeman) — Sat, 28 Jan 2012 10:03:12 +0000

Matt what was the host system when you ran the VM?

no-reply@allegro.cc (Matthew Leverton) — Sat, 28 Jan 2012 10:41:51 +0000

OS X 10.7.