In chapter 8.4.4.2.2, the substitution is described as following :

"The values of the samples p[ x ][ y ] with x = −1, y = −1..nT*2−1 and x = 0..nT*2−1, y = −1 are modified as follows: – If all samples p[ x ][ y ] with x = −1, y = −1..nT*2−1 and x = 0..nT*2−1, y = −1 are marked as "not available for intra prediction," the value ( 1 << ( BitDepthY − 1 ) ) is substituted for the values of all samples p[ x ][ y ]."

This is applied for luma and chroma pixels. For chroma pixels, I think the value to be used should be : ( 1 << ( BitDepthC − 1 ) )

Minor software issues related to SIFO: i) For B pictures directional filter is not checked properly. ii) Filter coefficients for directional is not switched as in separable filter.

In section 8.6.5 it is specified how the reconstruction is obtained. It is using predSamples and resSamples to obtain recSamples. However resSamples are not specified to be 0 when skip_flag[x0][yo] is equal to 1.

Not sure of the best place to indicate this. One possibility is in section of 8.5.4.

Perhaps something like what is added below between "" .. "": Depending on no_residual_syntax_flag, the following applies: – If no_residual_syntax_flag is equal to 1 ""or skip_flag[xC][yC] is equal to 1"", all samples of the (nCSL)x(nCSL) array resSamplesL and all samples of the two (nCSC)x(nCSC) arrays resSamplesCb and resSamplesCr are set equal to 0.