[gchpaco]

Lamport's papers are exceptionally readable, I should note. His discussion of e.g. Paxos is still current and interesting and still has implications for distributed databases even though that algorithm is too expensive for frequently (like, order of magnitude hundreds or thousands of changes a second) changing data.

The Dragon is mostly old and hoary by today's standards; we frequently use GLR instead of LALR(1), it has no discussion of interesting modern developments like packrat parsers, there is next to no discussion of any part of the compiler that isn't parsing, etc. A product of its time, but not an edifice for the ages in the way that SICP is.

If you find a good stats/probability book, let me know; I've been looking for one for years and come up short. Russell and Norvig is ironically superior to most texts in the field.

[beagle3]

You're wrong about _any_ edition of the the Dragon.

The 0th edition (green dragon, Aho&Ullman) was mostly about parsing, but talked about everything. The 1st edition (red dragon, Aho,Sethi&Ullman) was about everything not including JITs, vectorization, GCs and parallelization. The 2nd edition (purple dragon, Aho,Lam,Sethi&Ullman) is about everything, and they even dropped operator precedence (Pratt) parsing to make room for more non-parsing stuff.

[kenjackson]

there is next to no discussion of any part of the compiler that isn't parsing

Are you sure? There are 200 pages that cover lexing and parsing -- of a 950 page book.

If anything I'd argue that they've largely punted on parsing given that it's a lot less important than it used to be (which may explain why they didn't add GLR or packrat parsing) and focused on things that were considered more interesting issues in compilers.

[pbiggar]

I agree. I hate the Dragon book, and I particularly hate that it's the go-to book for learning compilers. It should not be. I recommend "Engineering a Compiler" by Cooper/Torczon, or "Modern Compiler Design in X" by Appel.