Out with old, in with the new.
If you had unlimited storage, you could keep everything forever. But you don't. At some point, you'll want to discard deleted items, and older versions of items, to make room for new changes. This is accomplished by merging layers.
Merging layers combines the changes captured in multiple layers into a single layer. During this process, multiple versions of items and deleted items are purged from the archive, retaining only an aggregate of the changes made during that time period.
The most common way to merge layers is to schedule a rolling merge action to be performed automatically.
If you're used to using other incremental backup utilities, the term “merge” might be confusing.
If you're familiar with source control systems, like git, then merging is the equivalent of "squashing" multiple commits into a single commit.
If you are used to thinking in terms of “deleting” or “collapsing” the second backup/session/snapshot/delta, that's OK. It doesn't matter if you think of merging as “combining layers 2 and 3” or as “deleting layer 2”, the result is the same.
Borrowing the example from the introduction to layers, let's see what happens when layers 2 and 3 are merged:
The new layer formed by merging layers 2 and 3 contains the changes made to document B, and a final version of document C. The original version of document C can no longer be recalled. (The original versions of documents A and B are still preserved in layer 1.)
Finally, let's merge the two remaining layers:
The resulting layer contains only the final versions of documents B and C. All trace of document A has been forgotten, along with those earlier versions of B and C. At this point, it is only possible to recall the last captured version of documents B and C.
Here's the simplest way to think of what happens when you merge layers:
The results of merging layers is identical to having never performed the earlier captures.
Stated another way, if you merge layers 5 through 7 in your archive, the single resulting layer will be exactly the same as if you had never performed the captures that created layers 5 and 6. You now have a single layer containing all of the changes that were captured in layers 5, 6, and 7—just as if they had all be captured at once.
The changes recorded in layers are not pure deltas, but positive deltas. It's a subtle distinction, but it's important for the robustness of the archive.
In the example above, layer 3 doesn't literally record that document A was deleted. It reasserts that document B still exists (unchanged) and records the changes to document C. The fact that document A was deleted is implied by its omission.
Should the information contained in layers 1 or 2 become lost for some reason (hardware malfunction, corrupted volume structure, and so on), an archive built on pure deltas would mean that layer 3 (and every layer after it) would be incomplete, since every layer would depend on the information recorded in the previous layers.
By using positive deltas—which subtly reassert key information in each layer—layer 3 can be reconstructed by the repair command, even in the absence of layers 1 and 2; it, and all subsequent layers, will be whole.