Conversation
It is one of the most common channel orders other than Rgb and the
construction involves nothing complicated, just swapping the channels
around. We don't support that layout during computation, where it would
complicate significant parts of `Pixel`, but as a helper for transferring
a buffer we can easily have both on the `Rgb{,A}` typed containers.
|
Do you know whether the trait bounds would be suitable if we wanted to replace the implementation with one using |
|
Hm, alright, likely not. |
|
Actually, if we made |
|
From what I understand, yes. The bound would be extremely weird but it seems feasible under the current interface. It does not seem very controversial to even expose I'm not entirely sold on sealing the |
These new bounds ensure that only our own pixel types can be used. That prepares changing our approach of algorithms such that we can optimize for individual cases without falling prey to needs for specialization. More context can be found here: <#2596 (comment)>
These new bounds ensure that only our own pixel types can be used. That prepares changing our approach of algorithms such that we can optimize for individual cases without falling prey to needs for specialization. More context can be found here: <#2596 (comment)>
2 participants
It is one of the most common channel orders other than Rgb and the construction involves nothing complicated, just swapping the channels around. We don't support that layout during computation, where it would complicate significant parts of
Pixel, but as a helper for transferring a buffer we can easily have both on theRgb{,A}typed containers.See: #2595 which seems like a paper cut that can be easily and confidently solved.