This sounds like something I ran into in my D lexer project. Lexing
Phobos took upwards of 10-15 seconds, which is extremely slow
considering that dmd can *compile* it in just a few seconds. So I did
some profiling, and found out that most of the time was spent formatting
tokens into strings and running writeln. After I commented that out, the
running time dropped to just a few seconds. :-)