The input is 20,000 single digit numbers. I would need at most 200KB to create a block map. It's worth a try and runs fast enough. I could have tracked each group of consecutive blocks instead, but I didn't feel the added complexity was worth it, yet.
First solution (puzzle2_slow.rb) took 5m14s. It moved backwards from the last file and looked for a free space large enough to accommodate it. O(n2) style.
On top of that, a file might get visited more than once. This creates extra work, but does not change the result. If the file didn't fit anywhere left of here before, it still won't fit anywhere left of here now.
One quick optimization is to stop looking when the file starts before the first free block, and to stop looking for free space past where the file starts. It is still O(n2) but runs in about 1m40s. But my two attempts to make it happen failed to give the correct checksum with the input data. The sample was fine, but something in the real inputs is throwing me off.
A better solution (puzzle2_fast.rb) is to track sequences of identical blocks. This makes finding free space for a file much faster, but requires splitting a sequence of free space into a sequence for a file and a possibly empty sequence of free space whenever we move a file. This has a complexity closer to O(n) and runs in 6s.