Hello,

I want to bring up an issue again about PyTorch dependency resolution on Intel-based macOS. This follows the discussion in #9515, which I wasn’t able to reopen.

I really appreciate the updates with the fork-strategy option and the closure of issues #8686 and #7190, but the main problem I brought up is still there: the dependency resolver enforces a single version of PyTorch across all platforms, even when conditional version constraints are specified in pyproject.toml.

Here’s a minimal example of my configuration:

[project]  
name = "test"  
version = "0.1.0"  
description = "Add your description here"  
readme = "README.md"  
requires-python = ">=3.11"  
dependencies = [  
    "torch>1.13,<2.3; sys_platform == 'darwin' and platform_machine == 'x86_64'",  
    "torch>1.13; sys_platform != 'darwin' or platform_machine != 'x86_64'",  
]  

[tool.uv]  
fork-strategy = "requires-python"  

The expectation here is that the resolver should pick PyTorch version 2.2.2 for Intel MacOS, and a different version (e.g., 2.5.1) for other platforms. However, the current behavior enforces a single version (2.2.2) across all platforms.

I was hoping the introduction of the fork-strategy option would address this limitation, but it doesn’t seem to allow opting out of this unified resolution behavior.

Would it be possible to consider adding a new fork-strategy option (or another mechanism) to handle cases like this? This would allow for platform-specific resolution of dependencies as defined in the pyproject.toml.