As a programmer who didn't take the compiler class in school, what's the best way to go about learning what it takes to write one? A book recommendation? Find a CS dept nearby and take the class?
_Essentials of Programming Languages_ (EoPL) is a great book about writing interpreters in Scheme, and the last chapter shows how to turn an interpreter into a continuation-based compiler. (It's in the first edition, I heard it was cut from the second and third as too advanced.)
Some people will recommend the Dragon book (_Compilers: Principles, Techniques, and Tools_ by Aho, et. al), which is great, if you want a dry book about writing a pre-modern language with clunky, old tools.
As a sweeping generalization, compiler books using Lisp, ML, and other languages with good support for symbolic computation get into more interesting material, because compiler books using e.g. C tend to spend a good chunk of the book on parsing and other issues that can be avoided entirely by working directly on trees in Lisp/ML/Prolog/etc.
Theory: Essentials of Programming Languages.
A Good Start: Programming Languages Application and Interpretation (Free online :-)
The Dragon book is probably the best introductory text. The new version that came out a few years ago actually did improve the original text a fair bit. Although there is a lack of discussion about SSA in the text.
While most compiler books spend most of the text talking about the front-end of the compiler the crazy interesting stuff is in the backend. The new Dragon book greatly improves here over the old one, but I'd also recommend two other texts:
1) Morgan, Building an Optimizing Compiler -- some people love the writing style. Others hate it, but the material is solid. 2) Muchnick, Advanced Compiler Design and Implementation -- Great book. Must read if you're serious about optimizations.
Computer Architecture: A Quantitative Approach, 4th Edition by Hennessy and Patterson. the Patterson and Hennessy is a good book to start with, too.
My knowledge is way out of date, but I think even back in 2000 those books were a bit behind the times in terms of optimizations production compilers do. For instance, there was little in-depth treatment of hot-cold optimizations in the 2nd edition.
The tension between processor architects and compiler writers is a constant theme.
It's like 90% of the practical knowledge of what a compiler does that I've ever needed to know for day-to-day coding.
However, I would recommend chapter 2, which builds a complete compiler for a very simple language (arithmetic expressions). In the first edition, that did it in C, and it's very simple and clear. In the second edition, they did it in Java, and making it object-oriented makes it much less clear. I therefore recommend the 1st ed (plus, the hand-drawn dragon on the cover looks much better than the purple cgi dragon on the 2nd ed.)
I have to agree that the Java code is an academic object-oriented programming example. It's not "less clear" so much as "less practical". For comparison, see the ANTLR API reference and the interfaces provided there, which differ just enough to show practical consideration for the use cases one might want.
The reason why I like this book is that it has a gentle slope for the reader but also discusses pragmatic ways for how you can increase confidence that (a) you implemented it correctly (b) the users of your compiler trust the code does what it says it does [such as good error messages essential to a compiler for an IDE]
I have lots of compiler books, too, including all editions of all Aho/Ullman books (going back to their pre-lex/yacc books that used the string processing language SNOBOL). This doesn't mean the others are bad or hard to read. On the contrary, you need to pick up skills over time from somewhere, and I usually pick them up from other books on compilers or LNCS's on compilers, etc.