Some incoming edges may have not their buffer size downsampled yet (but at least one has it thanks to the dfs ordering)
Some incoming edges could have not had their buffer size with the normal sample size yet if an incoming branch was downsampled (but at least one has it thanks to the dfs ordering).
But actually this case does not happen, as if the path is downsampled, then thanks to dfs ordering, all the subsequent edges including the incoming one of the current node, must have been explored.
If it had happened, we would have taken max in the case of ratio < 1.0
Some incoming edges may have not their buffer size downsampled yet (but at least one has it thanks to the dfs ordering)
Some incoming edges could have not had their buffer size with the normal sample size yet if an incoming branch was downsampled (but at least one has it thanks to the dfs ordering).
But actually this case does not happen, as if the path is downsampled, then thanks to dfs ordering, all the subsequent edges including the incoming one of the current node, must have been explored.
If it had happened, we would have taken max in the case of ratio < 1.0