Find a way to make sphinx output in tests more predictable

[flying-sheep]

The Sphinx output changes depending on if PEP 604 Union syntax is used. It’s not about from __future__ import annotations, it’s about actually using the syntax!

Even on versions < 3.10, a simple type like int gets compiled to different things, depending on what syntax is used in the file:

• "int" if the file contains no new-style union syntax,
• *int* if the file contains new-style union syntax,

E.g. when you don‘t use | union syntax,

from __future__ import annotations

def function(x: bool) -> str: ...

gets compiled to ":

...
Parameters:
   * **x** ("bool") -- foo

And when you do use | union syntax anywhere in the file (or is it just anywhere in the compilation unit?),

from __future__ import annotations

def function(x: bool) -> str | int: ...

gets compiled to *:

...
Parameters:
   * **x** (*bool*) -- foo

[gaborbernat]

A PR for this would we welcome.

[flying-sheep]

Idk what the best approach is, or even where to fix this. Can we configure Sphinx to produce more predictable output? Or do we need an upstream change there?

[gaborbernat]

That would be to discover :thinking: