It changes the content of the commit (by iterating the hash in the commit message itself) until the hash of the child commit matches. A big part of the trick is that it only looks at a prefix of the hash, so the search space is much smaller than the full SHA1 hash.
For anyone curious, this uses the same idea as the "guess-and-check" approach spullara mentioned: https://news.ycombinator.com/item?id=953569
It changes the content of the commit (by iterating the hash in the commit message itself) until the hash of the child commit matches. A big part of the trick is that it only looks at a prefix of the hash, so the search space is much smaller than the full SHA1 hash.