#[new] can silently fail when arguments to the function can't be constructed

[Walter-Reactor]

Bug Description

When using #[new] on a function where one of the parameters can't be constructed by PyO3, the constructor silently fails and attempting to construct the object will at runtime trigger a "No constructor defined" exception.

Steps to Reproduce

The following compiles with no warning or error, but attempting to create a Foo from python will fail.

#[pyclass]
#[derive(Clone)]
pub struct Param(String); // Simple! forgot to provide an impl block with new on it.

#[pyclass]
pub struct Foo{...}

#[pymethods]
impl Foo{
  #[new]
  pub fn new(param : Param) -> Self {
    Foo{}
  }

Backtrace

No response

Your operating system and version

Windows 11

Your Python version (python --version)

Python 3.12.4

Your Rust version (rustc --version)

rustc 1.80.1 (3f5fd8dd4 2024-08-06)

Your PyO3 version

0.22.1

How did you install python? Did you use a virtualenv?

System python, invoked via uv 0.3.0's uv run

Additional Info

No response

[attack68]

This has a crossover with #4465 which defines #[new] on simple enums for a separate reason.

[davidhewitt]

@Walter-Reactor can you provide a more complete minimal repro along with the proposed change to behaviour you would like to see?

The problem, as I see it currently, is that you cannot create a Param() because you didn't define a way, and Foo has nothing to do with this case.

[Walter-Reactor]

That's accurate, but the inability to construct Foo seemed like it was being contagious. Now that I consider it, if there were some other way to construct Param, that's not actually true.

My main issue here is one of diagnostsics. It was extremely confusing when I tried to write Foo("some string"), and was told that the constructor didn't exist, yet there was clearly a #[new] block on it. If the error message included the name of the type, I think this would be much clearer

[birkenfeld]

It was extremely confusing when I tried to write Foo("some string"), and was told that the constructor didn't exist

I can't reproduce this. Using the code you posted above:

>>> Foo("a")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: argument 'param': 'str' object cannot be converted to 'Param'
>>> Param("a")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: No constructor defined

Still, it could indeed be good to mention the type name in the "no constructor defined" message.