So? It's not that uncommon for highly dynamic languages to do this. Lisp macros for example depend on being able to run arbitrary code at compile-time.
Is it? (I honestly have no idea about Lisp.) It isn't just about being able to do stuff at compile time: In Perl's case, as I understand it, you can't determine if a string of text is a valid (in the sense of generating an AST!) Perl program without executing it. This precludes things like reliable syntax highlighting.
Languages can include meta-programming/macro facilities without effecting whether or not the parser can do its job; I am fairly certain that Rust's macros are an example of this. Python, as a more dynamic language, also includes metaprogramming facilities (metaclasses, decorators) but IIRC, can still be parsed without needing to simulate a Turing machine.
In Lisp you can modify the read table at run time, so you can do whatever you want. The functions the reader calls when it comes across '(', ')', ' ', and every other character are all completely customisable.
Well, for one thing, it really shortens the discussion if you take advantage of it. For all languages that require arbitrary code execution in that language at compile time in order to be parsed, parsing therefore is Turing complete and therefore has all properties of Turing completeness, including undecidability. QED.
