cool, thanks for checking. the problem seems to be pr 699 and afterwoes.. the core problem is a race condition on dt_image_t data in the pipeline: some more recent and more complicated raw tricks (some specific to esoteric sensors, some just about black/white levels) require rawspeed to load some metadata before the pipeline inits itself in reload_defaults or commit_params. this isn't always guaranteed, as the raw is loaded on demand in the thread for the full pipeline, and reload_defaults is a gui thread callback.
in fixing this (we now load the bloody raw first thing when entering darkroom mode, in the gui thread :( we could just load the metadata if there was support for this in rawspeed), we discovered that reload_defaults is called quite a few times, redundantly. but cleaning that up seems to be not just a matter of not calling it, but also teaching individual modules to work well with the calling order gui_init - reload_defaults - gui_update. especially colorin.c seems to have problems with that. now in master is the fix for the race condition, and nothing related to reload_defaults cleanup work. this means it's being called a ton of unnecessary times, and i have no confidence in the calling order for multi-instance modules. but this is the same situation as it was before, so that should be stupid from a dev point of view but not dangerous from a user's perspective. cheers, jo On Thu, Oct 23, 2014 at 6:58 AM, Ulrich Pegelow <ulrich.pege...@tongareva.de > wrote: > Hi Jo, > > the reported issue seems to be fixed with current master (commit > 164ad8cae3b40017e48581c143084fa3759405b7). I only checked the behavior of > the demosaic module. Please tell me if you need further testing from my > side on other modules. > > @Roman: > I noticed the problem with c15f26a6c57fbf5583a98f4de4c2b9d500751fb1 for > the first time. I tentatively moved back one commit but that didn't solve > the issue. > > Ulrich > > Am 22.10.2014 um 23:01 schrieb johannes hanika: > > sorry, i'm too tired to fix this for real tonight. this reload_defaults >> calling order seems to have a ton of non-transparent implications.. >> reverted a few cleanup commits for now. note that this re-introduces >> some problem with input colour profiles when switching images via film >> strip, if you're switching between jpg and raw and files that come with >> embedded profiles. maybe it needs to be fixed in colorin.c instead. >> >> let me know if that version works for you now. >> >> -jo >> >> >>
------------------------------------------------------------------------------
_______________________________________________ darktable-devel mailing list darktable-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/darktable-devel
