a property of the language, not the machine

This is the fine print of Rice's theorem, and getting it wrong leads people to wildly over-claim. The distinction is between WHAT a program does and HOW it is written. A property of the language (a semantic property) asks about the set of inputs the program accepts — its behaviour. A property of the machine (a syntactic property) asks about the program's source text — its structure. Rice's theorem condemns the first kind to undecidability; it says nothing against the second kind, many of which are perfectly decidable.

Think of two programs that both compute the exact same thing but are written completely differently — one in ten lines, one in a thousand, perhaps in different styles. They recognize the SAME language, so they share every semantic property: same accepted strings, same emptiness, same finiteness. But they differ in syntactic properties: different line counts, different variable names, different numbers of states. Rice's theorem is a statement about the things the two programs share (their common language), not about the things that distinguish their texts. 'Does L(M) contain hello?' is semantic and undecidable; 'does M have exactly 7 states?' or 'does M's source contain a goto?' is syntactic and decidable by simply inspecting the code.

The practical upshot saves you from a common error. People sometimes 'disprove' Rice's theorem by pointing out that a linter can detect, say, unreachable code or whether a function has three arguments — but those are syntactic facts about the text, fully decidable, and never claimed undecidable. Rice only bites once your question is really about behaviour: what outputs the program produces, what inputs it accepts, whether two programs behave alike. The boundary between syntax and semantics is exactly where the decidable and the undecidable part ways.