What code path actually ends up here with !seat? the only two call sites I'm aware of aren't capable of passing NULL.

I think we could put an EINA_SAFETY check at the top of ecore_drm2_device_open for seat, and just not allow it to be NULL ever.

Do you mean
EINA_SAFETY_ON_NULL_RETURN(seat);
?

In this case I think
EINA_SAFETY_ON_NULL_RETURN_VAL(seat, NULL);

But that takes us to this:
!seat is always false, and probably always has been

so the conditional in your change is always false, and we never filter out seat0 tagged devices from other seats?

Hm... ...I don't know what the second if statement was for...
...at first I thought that in case if dseat was NULL (in case if the seat attribute was not set in udev)

...but looking that again, I just realized the order of dseat is flipped in the second one... ...AND dseat is defaulted to seat0 it is null a few lines before, to account for udev returning null on the devices seat attribute...
so it seems that I guess if

_drm2_device_find

was called without the seat argument as NULL, assume that the caller meant the default "seat0"

I thought that maybe there was the possibility that calling _drm2_device_find with the second argument as NULL was valid in some way... (since it's pretty similar in ecore_drm_device_find in non v2 ecore_drm_device.c )

...if not I'll change it

I updated the diff
I don't know if ...is it valid for the first argument of ecore_drm2_device_open to be NULL? (assuming it is running on seat0 , otherwise it won't work)
because if it is this might be an API break compared to the last one I submitted...