I recall the classic 1986 edition of the Dragon Book spending waaaay too much time on boring details of lexing and parsing. I think it turned many people off the interesting aspects of compilers. It also had an interval-based formulation of data-flow analysis that nobody uses, possibly because it's (a) more complex than worklist iteration, and (b) not applicable to irreducible control flow graphs.

Are newer editions better and more focused on interesting solutions to interesting problems?

Try the Tiger Book instead (a.k.a. 'Modern compiler implementation in C', or there are Java or ML versions if those are more to your taste). It has a better balance: more to the point treatment of lexing and parsing, leaving more space for a more comprehensive treatment of optimization and code generation.

I second it, even if the book has a few issues here and there, it was one of the best books I read on the subject.

I recommend making sure to avoid reading this book, it spends all its time making mountains out of molehills in the easy parts and leaves out the hard parts. How did they write so many pages about text parsing?

The books by Appel and Louden (forgot the names) are good.