How does data movement work?

[timuckun]

You say that the code can reference other functions. Say I take the above 3 gigabyte image and break it to smaller chunks (which I believe IPFS does) and then write a wrapper function which refers to all the chunks. When somebody accesses my code how does it pull all the sub data to my workstation so that my data can be assembled in proper order. How do I include metadata so know how the tiling is supposed to work?

[dborzov]

Each kolmoblock has a header that includes meta information such as what dependency blocks need to be resolved in advance as they are "imported" when the block is executed. The header contains all the meta information, so the client can get it without downloading all the data and make a decision on whether it wants to pursue it.

It's a clients responsibility to figure out where to get the dependency blocks.