Parser Generators Considered Useless?
alumnit.ca
alumnit.ca
It's actually common for people to use a parser generator when implementing a compiler for a language. He's looking at the tip of the iceberg - quality compilers for popular languages - and assuming that's all there is.
If you're by yourself or on a small team, it's a waste of your time to build a parser for a language by scratch. You might be able to produce a better-performing parser if you implement it yourself, but the performance of your compiler isn't even a second-order concern, it's a third or fourth order concern. I say this because you're going to care about the performance of the code you generate far before you care about the performance of the compiler itself. On par with that is, of course, language features. If your language features require a change in the grammar, it's much easier to change a grammar file than to muck around in your own implementation.
Popular compilers do take time to optimize parsing, but very few compilers have to handle the strain that the likes of gcc and MSVC do.
Unfortunately, that seems to be where my formal education ends, because I just can't figure out why lexical analysis is supposed to be so difficult.
It's not difficult. It's just tedious. Which is why the formal theory exists: given the description of all tokens in a language, we can generate a state machine which will tokenize the input stream.