- Psalm is slow in very large monorepos (over 5 million lines of checked-in code).
- Psalm’s language server is more limited in what it can provide than comparable compiled-language tools. For example, it's infeasible to find/change all occurences of a symbol, in all files that use it, as you type.
Python is a special case where the semi-official typechecker [MyPy](https://github.com/python/mypy), written in Python, can also [compile to a C Python extension](https://github.com/python/mypy/blame/master/mypyc/README.md#L6-L10), which runs 4x faster than the interpreted equivalent.
Some interpreted languages have unofficial typecheckers written in faster compiled languages. In all cases a single mid-to-large company is behind each effort, with a small number of contributors not employed by that company:
Psalm is mostly run on PHP code that parses a lint check (`php -l`). Psalm does not replace that check for verifying PHP syntax.
Since most of the time Psalm is used on syntatically-correct code it should use a parser built for that purpose, and `nikic/php-parser` is the best candidate.
Psalm offers a plugin API that allows you to tell it what about your program's property types, return types and parameter types.
Psalm aims to operate in a space with other static analysis tools. The more users rely on plugins, the less chance those other tools have to understand the user's intent.
So wherever possible Psalm should encourage developers to use annotations rather than type-providing plugins. This was a driving force in the adoption of [Conditional Types](../annotating_code/type_syntax/conditional_types.md) which allowed Psalm to replace some of its own internal type-providing plugins.
