Issue with .whl extension in dependency. #7284
Description
Following on from my questions in #7245, trying to pin pytorch for multiple architectures.
[edit] Have retried with 0.4.9 release and get same warnings. [/edit]
Minimum reproducible pyproject.toml
[project]
name = "uv-test"
version = "0.0.0"
requires-python = ">=3.10,<3.12"
dependencies = [
"torch @ https://download.pytorch.org/whl/cpu/torch-2.0.1-cp310-none-macosx_11_0_arm64.whl ; sys_platform == 'darwin' and platform_machine == 'arm64'",
"torch @ https://download.pytorch.org/whl/cpu/torch-2.0.1-cp310-none-macosx_10_9_x86_64.whl ; sys_platform == 'darwin' and platform_machine == 'x86_64'",
"torch == 2.0.1+cpu ; sys_platform != 'darwin'",
]
[tool.uv]
extra-index-url = ["https://download.pytorch.org/whl/cpu"]
When locking on linux, all goes well
>> uv lock
Using Python 3.11.9 interpreter at: /usr/bin/python3.11
Resolved 11 packages in 1.22s
When syncing on linux, there is a warning (the root of my question), but installation continues successfully.
>> uv sync
Using Python 3.11.9 interpreter at: /usr/bin/python3.11
Creating virtualenv at: .venv
warning: Failed to validate existing lockfile: failed to parse file extension; expected one of: `.zip`, `.tar.gz`, `.tar.bz2`, `.tar.xz`, or `.tar.zst`
Resolved 11 packages in 754ms
Installed 8 packages in 79ms
+ filelock==3.13.1
+ jinja2==3.1.3
+ markupsafe==2.1.5
+ mpmath==1.3.0
+ networkx==3.2.1
+ sympy==1.12
+ torch==2.0.1+cpu
+ typing-extensions==4.9.0
However when syncing on a mac.
>> uv sync
Using Python 3.11.9 interpreter at: /opt/homebrew/opt/[email protected]/bin/python3.11
Creating virtualenv at: .venv
warning: Failed to validate existing lockfile: failed to parse file extension; expected one of: `.zip`, `.tar.gz`, `.tgz`, `.tar.bz2`, `.tar.xz`, or `.tar.zst`
Resolved 95 packages in 4.13s
error: failed to parse file extension; expected one of: `.zip`, `.tar.gz`, `.tgz`, `.tar.bz2`, `.tar.xz`, or `.tar.zst`
Caused by: `.zip`, `.tar.gz`, `.tgz`, `.tar.bz2`, `.tar.xz`, or `.tar.zst`
Is this me failing to use uv correctly, or should a .whl be a viable source (lock command seems to think so)
[[package]]
name = "torch"
version = "2.0.1"
source = { url = "https://download.pytorch.org/whl/cpu/torch-2.0.1-cp310-none-macosx_10_9_x86_64.whl" }
resolution-markers = [
"platform_machine == 'x86_64' and sys_platform == 'darwin'",
]
dependencies = [
{ name = "filelock", marker = "platform_machine == 'x86_64' and sys_platform == 'darwin'" },
{ name = "jinja2", marker = "platform_machine == 'x86_64' and sys_platform == 'darwin'" },
{ name = "networkx", marker = "platform_machine == 'x86_64' and sys_platform == 'darwin'" },
{ name = "sympy", marker = "platform_machine == 'x86_64' and sys_platform == 'darwin'" },
{ name = "typing-extensions", marker = "platform_machine == 'x86_64' and sys_platform == 'darwin'" },
]
wheels = [
{ url = "https://download.pytorch.org/whl/cpu/torch-2.0.1-cp310-none-macosx_10_9_x86_64.whl", hash = "sha256:567f84d657edc5582d716900543e6e62353dbe275e61cdc36eda4929e46df9e7" },
]