Ulli wrote:
Lubomir wrote:
Do you see any issues with the GIL interfaces that would prevent porting the Vigra algorithms to GIL, or that would affect their performance? If yes, what interfaces are missing and what algorithms need them?
Three things come to mind: - many more type conversions; e.g. int8 => uint16; RGB <=> LAB etc. - adaptors for the subsampled image view - support for multi-dimensional images (e.g. volume data, image sequences)
My question was not what could be added to GIL, my question was what in your opinion needs to _change_ in the GIL interfaces and design, if anything. Adding new functionality in later versions of the library is perfectly normal, and I foresee additions to the GIL core once the algorithms are implemented. Nothing you mentioned requires changes (the PixelAccessor issue is discussed in another thread, and even for it we agree it is possible to have adapters) Lubomir