14:24 <apritzel> Hi, I need some help with a nasty circular Kconfig dependency issue, what would be the best approach? Send a (RFC) series exposing the problem, or send a separate error report message to the list? Or describe and discuss that here?

14:26 <milkylainen> What part of things? kconfig maintainers? Or is it related to something structural in an arch?

14:26 <apritzel> the symbol triggering this is CONFIG_DRM_PANEL_EDP, which misses to select CONFIG_DRM_KMS_HELPER, but just naively adding a select doesn't work

14:26 <apritzel> the typical Kconfig dependency rabbit hole

14:27 <apritzel> arnd: ^^^ this came up when fixing multi_v5_defconfig

14:28 <apritzel> milkylainen: so it's just one nasty instance, probably more of interest for the respective subsystem maintainers

14:29 <milkylainen> apritzel: I was thinking subsystem yes.

14:35 <arnd> apritzel: does https://lore.kernel.org/dri-devel/20220315084559.23510-1-tzimmermann@suse.de/ help?

14:35 <arnd> circular dependencies happen when you mix 'depends on' and 'select'

14:35 <arnd> in the case of CONFIG_DRM_KMS_HELPER, I think everyone should use 'select'

14:36 <arnd> see also https://lore.kernel.org/lkml/CAK8P3a3ogV8sJshQJrzHe7hc2k1Opq-2hvXf6DU+HfhREk-DhQ@mail.gmail.com/

14:37 <apritzel> arnd: thanks for the pointer to the patch, I will give this a try

14:38 <apritzel> and yeah, I tried the solutions "by the book", but gave up after some hours :-(

15:35 <apritzel> arnd: many thanks for the pointer, that indeed seems to fix it for me!

15:35 <apritzel> funny how this problem seems to be quite long standing, but the patch is two days old ;-)

15:42 <javierm> apritzel: maybe it was exposed by commit adb9d5a2cc77 ("drm/dp: Move DisplayPort helpers into separate helper module") that's fairly recent ?

15:43 <javierm> the fact that from the fix mentioned by arnd you can follow a very long Fixes: chain shows that these kind of Kconfig deps issues are quite hard to solve correctly

15:43 <javierm> *from the fix

15:43 <apritzel> javierm: I think this might be the trigger for that patch, but that problem seemed to be around for much longer

15:44 <apritzel> javierm: some other problem made this symbol to disappear from the multi_v5_defconfig, so nobody notices that

15:45 <apritzel> javierm: and other configs still had this symbol, but also something else that selected the helper, so again that worked

15:45 <javierm> apritzel: yes, agreed. I just meant that splitting into a separate module exposed it since before both object files were in the same module