I agree that it's kinda dirty, but I don't think there's a way to access the "raw" data of the #[MyAttrib(...)] "annotation" otherwise? Like, instead of actually constructing an object just statically parse the annotation data. Quite similar to what getArgumentsExpressions() already seems to do, but there you're still missing out on the argument names. Perhaps phpstan could follow the attribute to its constructor and extract the names from there? Then all that's left is a way to get/reconstruct the actual value. I don't think there are any runtime values in this scenario? Of course the constructor could still modify the passed arguments before actually storing them, but I don't think that's really any different from a regular (non-attribute) class doing some stuff in its constructor? As far as I know phpstan is at least somewhat aware of values, like having @var 'foo'|bar' and trying to set 'baz' should result in an error. I'm not entirely sure what would happen if an attribute accepts an array and the constructor transforms this array, though.

I did indeed find both those links, but none of the docs from the first link seem to mention attributes at all. The second link is pretty much just the code and accompanying doc blocks presented in a different format, which doesn't always (clearly) explain what certain functions are used for or how to use them. Example: I found another getAttributes() function and since it's in the context of a "node", I figured it could be the type of attribute I'm talking about. But apparently it's something entirely different and it's more like PhpParser's internal attributes for that node? Or this function, which initially didn't do what I wanted but I passed class::method strings for the $constructors argument, while I should've passed just method. I'm not really seeing that type of information anywhere in the docs.