[Feature Request] Rename `[Unsafe]Pointer.address_of(x)` to something shorter

soraros commented 1 month ago

Review Mojo's priorities

[X] I have read the roadmap and priorities and I believe this request falls within the priorities.

What is your request?

As title. Maybe to Pointer(to=thing).

What is your motivation for this change?

I think they take too much line space
Quote from Chris: "In general, in my opinion, we should use initializers to create instances of a type, not static methods."
Again, from Chris: "so long as UnsafePointer and Pointer are consistent, I am happy"

Any other details?

For the bike-shading crowd: I mostly just moved Chris's comments from Discord, please don't shoot the passenger.

melodyogonna commented 1 month ago

+1 to this, conveys the same meaning with a more succinct syntax, definitely more Pythonic.

lattner commented 1 month ago

I don't think that making it shorter is the only motivation, it is also about consistency: I'd suggest using initializer syntax, so that "creating a T" is always done with initializers.

If we did that, it would argue for something like Pointer(to=x) or Pointer(at=x) or Pointer(of=x) or Pointer(address=x) etc.

martinvuyk commented 1 month ago

Yesss, this will even spare having to set the type that the pointer is for since it can be inferred :+1:

Pointer(at=x) [...] Pointer(address=x)

these ones in particular might be confusing if you pass an Int and you come from the C world though

martinvuyk commented 1 month ago

Opened PR #3606 but I'm getting a segfault for _ListIter even though I only added a constructor for Pointer and didn't touch any more code :/

soraros commented 1 month ago

If we did that, it would argue for something like Pointer(to=x) or Pointer(at=x) or Pointer(of=x) or Pointer(address=x) etc.

More broadly speaking, the fact that Mojo supports function overloads opens the door to Swift-style, argument-label-rich API designs. I'm overall supportive to this approach, but we also need to aware of the trade-offs.

For object initialisation, static methods (technically classmethods) are like a "poor man's" overloading in Python. They do have some advantages though:

They are more "Pythonic", at least in the sense that Python allows this approach (e.g., don't require hacks like those used in range to support 3 different arities).
This is likely not the case for Pointer, but it could be important for Python interoperability for some other types (especially when called from Python).
They also have the advantage of making the main argument not keyword-only, which makes them more compatible with higher-order functions.
They require less LSP support. The LSP doesn't show argument labels in auto-completion, and since it's part of an overload set, the LSP doesn't display the correct docstring, making them even more inaccessible.

modularml / mojo