Merkle DAG reconstruction on write ops

[richardhundt]

Hi guys, awesome project! I have a question though.

If the nodes in the DAG are stored as immutable content addressable blobs, surely if I write a single byte to a (logical) file somewhere down in the hierarchy, then because its content hash changes, anything which references it (tree node, for example) must also be updated, and it in turn will have a new hash, and so on all the way up to the root key after the mutable namespace?

In other words, every change anywhere in a tree will cause you to have to recursively recompute a new DAG for the ancestors. Isn't this kinda expensive? Think mounting a VM disk image where lots of writes are happening? I think this is why Ivy chose a log-based approach. Or have I missed something?

[EDIT] of course, I'm assuming "commit on file close after writing" semantics