Add an anti-pitch for typing

docs/source/guides.rst

@@ -9,3 +9,4 @@ Type System Guides
    libraries
    writing_stubs
    unreachable
+   typing_anti_pitch

docs/source/typing_anti_pitch.rst

@@ -0,0 +1,76 @@
+.. _typing-anti-pitch:
+
+Reasons to avoid static type checking
+=====================================
+
+In the words of :pep:`484`::
+
+    It should also be emphasized that Python will remain a dynamically typed language, and the
+    authors have no desire to ever make type hints mandatory, even by convention.
+
+The idea that dynamism in Python is a strength of the language is reflected in the fact that
+Python's type system is gradual. See :pep:`483` for details, but the long and short of this is
+that you can add static types to your codebase only to the extent that you want to, and static
+type checkers are designed to be able to put up with this.
+
+It's also worth noting that "static type checking" encompasses a spectrum of possible degrees of
+strictness. On the one hand, you can set yourself up so that your type checker does almost nothing.
+On the other -- well, I love type checking, but I would quit Python if I had to enable all
+possible strictness checks that type checkers offer.
+
+Anway, with all that said, here's a list of possible reasons to not use static type checking
+in Python::
+
+* You simply don't want to. Python is a tool that is meant to serve you. Python is a big tent,
+  multi-paradigm language that generally allows you to do things in the way that best suits your
+  needs, as best determined by you.
+
+* The cost-benefit ratio isn't good enough. Pleasing static type checkers requires a non-zero amount
+  of busy work. If this isn't worth the extra confidence you get, you shouldn't add static type
+  checking.
+
+* Your codebase fits in your developers' heads. Opinions vary, but people tend to agree that at
+  some number of developers and lines of code, static type checking confers significantly more
+  benefit. You don't feel like you're there yet.
+
+* If you maintain high test coverage, that might provide sufficient quality assurance for your
+  needs (acknowledging that static type checking and tests enforce different things; static type
+  checking usually cannot validate logic, tests can often not prove invariants of your code to
+  hold).
+
+* Your codebase is old, large and has been working fine without static type checking for years.
+  While Python's type system is designed to
+  `allow gradual adoption of static type checking <https://mypy.readthedocs.io/en/stable/existing_code.html>`_,
+  the total cost of adding type annotations to a large extant codebase can be prohibitive.
+
+* Your library does enough dynamic things that type checking would be unlikely to help your users.
+  Either you would have to spend a lot of effort redesigning your library in ways that static type
+  checkers could better understand, or you'd have to invest a lot of effort into figuring out clever
+  type annotations to twist the arms of type checkers.
+
+* Your codebase has suffered at the hands of `Hyrum's Law <https://www.hyrumslaw.com/>`_
+  and all possible observable behaviour is depended on. In order to avoid false positives for your
+  users, all your types end up being either a) complicated ``Protocol``s that are hard to maintain,
+  or b) ``Any`` in which case there's not much point. (On the other hand, static type checking could
+  be a good solution for communicating to users what behaviour they should be allowed to rely on)
+
+* You're not opposed to type checking in theory, but you dislike Python type checkers in practice.
+  Maybe they don't understand enough of the idioms you use, maybe you'd like them to infer more
+  instead of relying on explicit annotations, maybe they're too slow, maybe they don't integrate
+  well with your editor, maybe they're too hard to configure. Whatever the reason -- it just doesn't
+  work for your project.
+
+Advice for maintainers of untyped libraries
+*******************************************
+
+You've made the decision that adding static types isn't the right choice for your library. But
+perhaps you'd still like to help your users who do use static type checking (and maybe you have some
+enthusiastic would-be contributors willing to help with this).
+
+One option is to publish a :pep:`561` stub-only package that is maintained separately from your
+main project. You could also contribute these stubs to the `typeshed <https://github.com/python/typeshed>`_
+project.
+
+If more people pester you about adding static types, feel free to link them to this document. And if
+you ever change your mind, make sure to check out some of the other guides in this documentation,
+and ask any questions you have over at `Python's typing discussions <https://github.com/python/typing/discussions>`_.