If you don't have that exact book to hand, you swine, then the wikipedia page looks comprehensive from a glance, if a bit inaccesible.
The key mistake was underestimating the ability to automate and parallelize a brute-force search. Turns out a keyspace of 3^26 is just too small. User error was a large contributing factor, of course, and ensured that the implementation details were always leaked. But when it came down to it, the Polish and British intelligence services could simply break the codes on a near-daily basis.
The actual keyspace was around 76 bits , which is fairly respectable, particularly for the mid-20th century. Much of the weakness was, in fact, the result of the Enigma being an early device in a nascent technological field -- ie, the result of failures in the Enigma's design, as well as procedures surrounding its operation (not just operator error; for example, Polish cryptanalysts used the procedural repeating of the rotor setting initialization (it's conceptually similar to an initialization vector in modern crypto) to break early versions of the machine as early as 1932.
Pre WWII of course, they figured if a PHD mathematician couldn't crack it in a couple hours with chalk and chalkboard, well, surely they'll just give up, right?
The ratio of optimism to pessimism was a bit off.
Not much has changed then.
It is interesting to note that the actual key settings used the indicator 'SIG' with ring settings 'PMP', but the [different] recovered key gives an identical decryption.
So there are keys of the enigma that produce the same ciphertext? Wouldn't that reduce the practical keyspace dramatically?
Also, would that mean a message could be "hardened" by referring to the original key within it? (E.g. "Execute order <second letter of indicator><first letter of indicator>)
The thing that helped me most was building a paper enigma.