Describe the bug
make sync fails on any uv older than 0.9.25, and the error points at this repository's own files rather than at the tool version, so it reads as though the checkout is broken.
AGENTS.md and the README ask for "uv installed" without a floor, while CI pins 0.11.14 in every workflow. A contributor between those two points gets no usable signal.
Debug information
- Agents SDK version:
main at d40f5d9
- Python version: 3.13.11 (also reproduced on 3.12.10)
- Operating system: Windows 11
- Does the issue reproduce with the latest release? Yes, the two files involved are unchanged.
- Does it occur consistently or intermittently? Consistently, it is version-determined.
$ uv sync --all-extras
warning: Failed to parse `pyproject.toml` during settings discovery:
TOML parse error at line 233, column 36
|
233 | exclude-newer-package = { openai = false }
| ^^^^^
data did not match any variant of untagged enum Helper
error: Failed to parse `uv.lock`
Caused by: TOML parse error at line 11, column 1
|
11 | [options]
| ^^^^^^^^^
data did not match any variant of untagged enum Helper
Neither line is malformed. exclude-newer-package and the [options] table with exclude-newer-span are both newer uv features, and an older uv reports them as invalid TOML.
Repro steps
Bisected against this checkout, running uv sync --all-extras from a clean clone:
uv 0.9.21 uv.lock parse: FAIL
uv 0.9.23 uv.lock parse: FAIL
uv 0.9.24 uv.lock parse: FAIL
uv 0.9.25 ok -> Resolved 202 packages in 2ms
uv 0.9.30 ok
uv 0.10.0 ok
uv 0.11.14 ok <- what CI pins
So the floor is exactly 0.9.25. 0.9.24 was released about five weeks before it, so this is reachable by anyone who installed uv recently and has not updated, not only by someone on a very old version.
Expected behavior
Either of these would close it, and I have no preference:
Declare the floor so uv reports it itself. [tool.uv] required-version = ">=0.9.25" turns both errors into a message naming the installed and required versions.
Or state it in AGENTS.md and the contributing section next to the existing uv prerequisite, alongside the Python 3.10+ line that is already there.
The first is self-maintaining and fails at the right layer; the second costs nothing and is visible before the first command runs. Pinning the same version CI uses is a third option but seems heavier than the problem.
Happy to send a PR for whichever shape you prefer. Filing first because the choice is a project convention rather than a bug fix, and #4472 suggested I ask rather than guess on that kind of thing.
Describe the bug
make syncfails on anyuvolder than 0.9.25, and the error points at this repository's own files rather than at the tool version, so it reads as though the checkout is broken.AGENTS.mdand the README ask for "uvinstalled" without a floor, while CI pins0.11.14in every workflow. A contributor between those two points gets no usable signal.Debug information
mainat d40f5d9Neither line is malformed.
exclude-newer-packageand the[options]table withexclude-newer-spanare both neweruvfeatures, and an olderuvreports them as invalid TOML.Repro steps
Bisected against this checkout, running
uv sync --all-extrasfrom a clean clone:So the floor is exactly 0.9.25. 0.9.24 was released about five weeks before it, so this is reachable by anyone who installed
uvrecently and has not updated, not only by someone on a very old version.Expected behavior
Either of these would close it, and I have no preference:
Declare the floor so
uvreports it itself.[tool.uv] required-version = ">=0.9.25"turns both errors into a message naming the installed and required versions.Or state it in
AGENTS.mdand the contributing section next to the existinguvprerequisite, alongside the Python 3.10+ line that is already there.The first is self-maintaining and fails at the right layer; the second costs nothing and is visible before the first command runs. Pinning the same version CI uses is a third option but seems heavier than the problem.
Happy to send a PR for whichever shape you prefer. Filing first because the choice is a project convention rather than a bug fix, and #4472 suggested I ask rather than guess on that kind of thing.