## Summary
resolves https://github.com/astral-sh/uv/issues/3272
added it as a new subcommand rather than a flag on an existing
command since that seems more consistent with `cargo tree` + cleaner
code organization, but can make changes if it's preferred the other way.
Exposes the option added in #4416. Adds `--toolchain-preference` and
`tool.uv.toolchain-preference` to configure if system or managed
toolchains are preferred. Users can opt-out of managed toolchains or
system toolchains entirely as well.
Restores the `PythonEnvironment::find` API which was removed a while
back in favor of `Toolchain::find`. As mentioned in #4416, I'm
attempting to separate the case where you want an active environment
from the case where you want an installed toolchain in order to create
environments.
I wanted to drop `EnvironmentPreference` from `Toolchain::find` and just
have us consistently consider (or not consider) virtual environments
when discovering toolchains for creating environments. Unfortunately
this caused a few things to break so I reverted that change and will
explore it separately. Because I was exploring that change, there are
some minor changes to the `Toolchain` API here.
Adds support for the toolchain discovery preferences outlined in
https://github.com/astral-sh/uv/issues/4198 but we don't expose this to
users yet, I'll do that next to make it easier to review.
I've made some refactors in the toolchain discovery implementation to
enable this behavior and move us towards clearer abstractions. There's
still remaining work here, but I'd prefer tackle things in follow-ups
instead of expanding this pull request. I plan on opening a couple
before merging this.
I'd like to shift the public toolchain API to focus on discovering
either an **environment** or a **toolchain**. The first would be used by
commands that operate on an environment, while the latter would be used
by commands that just need an interpreter to create environments. I
haven't changed this here, but some of the refactors are in preparation
for supporting this idea.
In brief:
- We now allow different ordering of installed toolchain discovery based
on a `ToolchainPreference` type. This is the type we will expose to
users.
- `SystemPython` was changed into an `EnvironmentPreference` which is
used to determine if we should prefer virtual or system Python
environments.
- We drop the whole `ToolchainSources` selection concept, it was
confusing and the error messages from it were awkward. Most of the
functionality is now captured by the preference enums, but you can't do
things like "only find a toolchain from the parent interpreter" as
easily anymore.
Previously, distributions created through `Source::Workspace` would have
the absolute path as lock path. This didn't cause any problems, since in
`Urls` we would later overwrite those urls with the correct one created
from being workspace members by path.
Changing the order surfaced this. This change emits the correct lock
path. I've manually checked the difference with `dbg!`, this is not
observable on main, but on the diverging urls branch it fixes lockfile
creation.
When a fork is created from a list of dependencies, we were previously
adding all other sibling dependencies to every fork created. But this
isn't actually quite right, since the fork created is always created by
some marker expression. And while it is definitively disjoint from any
directly conflicting dependency specification, it is also possibly
disjoint with other dependencies. For example, as reported in #4414:
```toml
dependencies = [
"anyio==4.4.0 ; python_version >= '3.12'",
"anyio==4.3.0 ; python_version < '3.12'",
"b1 ; python_version >= '3.12'",
"b2 ; python_version < '3.12'",
]
```
The first two `anyio` requirements are conflicting with non-overlapping
marker expressions, and so a fork is created. Prior to this commit,
*both* `b1` and `b2` would be added to each fork. But of course, `b2` is
impossible in the `anyio==4.4.0` fork because of disjoint marker
expressions.
So in this commit, we specifically filter out any sibling dependencies
that could find their way into a fork that have disjoint markers with
that fork. We are careful to do this both when a new fork is created
from an existing set of dependencies, and when adding new dependencies
to a fork.
Fixes#4414
## Summary
After this change, `uv add` will try to use `tool.uv.sources` for all
source requirements. If a source cannot be resolved, i.e. an ambiguous
Git reference is provided, it will error. Git references can be
specified with the `--tag`, `--branch`, or `--rev` arguments. Editables
are also supported with `--editable`.
Users can opt-out of `tool.uv.sources` support with the `--raw` flag,
which will force uv to use `project.dependencies`.
Part of https://github.com/astral-sh/uv/issues/3959.
To support diverging urls, we have to check urls when adding
dependencies (after forking). To prepare for this, i've moved adding
dependencies for the current version to
`SolveState::add_package_version_dependencies` and removed the
duplication when checking for self-dependencies.
This changed is joined with a change in pubgrub
(https://github.com/astral-sh/pubgrub/pull/27) that simplifies the same
code path.
## Summary
Implements `uv add foo --workspace`, which adds `foo` as a workspace
dependency with the corresponding `tool.uv.sources` entry.
Part of https://github.com/astral-sh/uv/issues/3959.
Specifically, these are emitted when requirements fail to satisfy
`Requires-Python` or the markers associated with the current fork in the
resolver.
Closes#4373
Closes#4380
This is the same logic as `should_stop_discovery` but I changed the log
level and duplicated it because I don't really want that method to be
public. Maybe it should be though?
As in #4360, updates the uv project CLI to respect `.python-version`
files as default Python version requests. Additionally, updates project
interpreter discovery to fetch managed toolchains as in `uv venv
--preview`.
It was becoming problematic that the virtual environment test context
diverged from the other one i.e. we had to implement filtering twice.
This combines the contexts and tweaks the `TestContext` API and
filtering mechanisms for Python versions. Combined with my previous
changes to the test context at #4364 and
https://github.com/astral-sh/uv/pull/4368 this finally unblocks the
snapshots for test cases in #4360 and
https://github.com/astral-sh/uv/pull/4362.
## Summary
Make the git commit/date part of the version string matched in
cache_prune optional, so that the test also works correctly when uv is
built from an unpacked release tarball rather than a git repository.
Fixes#4374
## Test Plan
Ran `cargo test` inside the git repository and inside unpacked archive
for `0.2.12` release with the patch applied on top.
## Summary
Adds a `--no-clean` flag to `uv sync` that keeps extraneous
installations. This is the default in `uv run` and `uv add`, but not in
`uv sync` or `uv remove`. This means you need to run an explicit `uv
sync/remove` to clean the virtual environment.
Extends new filters for interpreter paths to apply to tests with
multiple Python versions. Adds patch version filtering for them as well,
which is needed for #4360 tests.
Extends #4364 automatically adding filters to the test context for
additional test virtual environments.
It turns out that the `pip sync` tests were really on the loose with
their virtual environment creation and it was difficult to use the new
helpers because they require mutability and the tests had immutable
borrows to extend the filters 😭. I refactored the tests to just reset
the environment.
A bare `uv toolchain install` invocation now reads default requests from
Python version files in the working directory. In order, a bare
invocation means:
- requests from `.python-versions`
- a single request from`.python-version`
- any installed managed toolchain
- the latest managed toolchain download
This replaces all the functionality of `cargo dev fetch-python`, which
we drop in #4337
## Summary
Closes https://github.com/astral-sh/uv/issues/4162
Changes keyring subprocess to allow display of stderr.
This aligns with pip's behavior since pip 23.1.
## Test Plan
* Tested using gnome-keyring-backend on a self-hosted private registry
as well as the keyring script described in #4162 to confirm both
existing functionality and the new stderr display.
* Existing tests using `scripts/packages/keyring_test_plugin` are now
showing its stderr output as well.
Allows installation of multiple toolchains in a single invocation
because I don't want to be limited to one! Most of the implementation
for concurrent downloads ported from `cargo dev fetch-python`.
Adds support for toolchain keys e.g. `cpython-3.11.2-macos` allowing you
to download toolchains for specific architectures and operating systems
using the format we use to uniquely identify a toolchain.
Closes#4240
e.g.
```
❯ cargo run -q -- pip install anyio --python "/Users/zb/Library/Application Support/uv/toolchains/cpython-3.12.0-macos-aarch64-none/install/bin/python3"
error: The interpreter at /Users/zb/Library/Application Support/uv/toolchains/cpython-3.12.0-macos-aarch64-none/install is externally managed, and indicates the following:
This toolchain is managed by uv and should not be modified.
Consider creating a virtual environment with `uv venv`.
```
I found this while testing the tracking of marker expressions across
resolver forks. Namely, given
sys_platform == 'darwin' and implementation_name == 'pypy'
And:
sys_platform == 'bar' or implementation_name == 'foo'
These should be disjoint, but the disjointness checker was reporting
them as overlapping. I fixed this by giving handling of disjunctions
higher precedence than conjunctions, although I am not 100% confident
that this is correct for all cases.
This commit adds marker expressions to our `Fork` type, which are in
turn passed down into `PubGrubDependencies::from_requirements` to filter
our any dependencies with markers that are disjoint from the fork's
marker expression.
This is necessary to avoid visiting packages in the dependency graph
that can never actually be installed. This is because when a fork is
created in the resolver, it always happens when there are two sibling
dependency specifications on a package with the same name, but with
non-overlapping marker expressions. Each fork corresponds to each
such conflicting dependency specification, and each fork assumes the
the corresponding marker expression as a pre-condition for any future
dependencies considered by it. That is, since the fork represents an
installation path that can only be taken when the corresponding
dependency specification (and its marker expression) is actually used,
it also therefore follows that the marker expression is true. Therefore,
any dependency visited in that fork with a marker expression that cannot
possibly be true when the markers of the fork are true can and ought to
be completely ignored.
There are some key invariants that I had to re-learn by reading the
code. This hopefully makes those invariants easier to discover by future
me (and others).
Otherwise, when testing `uv venv --preview` the default toolchain
directory will leak into the test.
I believe I've made a similar change for the standard `TestContext` in
another commit somewhere in my stack, if not I'll add it after.
Adds a command to find a toolchain on the system. Right now, it displays
the path to the first matching toolchain. We'll probably have more rich
output in the future (after implementing `toolchain show`).
The eventual plan (separate from here) is to port all of the toolchain
discovery tests to use this command. I'll add a few tests for this
command here anyway.
## Summary
This would be a lightweight solution to
https://github.com/astral-sh/uv/issues/4307 that doesn't fully engage
with all the possibilities in the design space (but would unblock
cross-platform for now).
## Summary
The fixtures here are pretty large, but it lets us test what we actually
care about (the resolved settings) rather than inferring the resolved
settings from behavior, which I think is a big improvement.
I also broke the tests down into more granular cases.
## Summary
Because the workspace member itself is part of the resolution, adding
the workspace name for the project leads to confusing errors, like:
```
❯ cargo run lock --preview
Compiling uv v0.2.11 (/Users/crmarsh/workspace/puffin/crates/uv)
Finished `dev` profile [unoptimized + debuginfo] target(s) in 1.79s
Running `/Users/crmarsh/workspace/puffin/target/debug/uv lock --preview`
× No solution found when resolving dependencies:
╰─▶ Because only albatross==0.1.0 is available and albatross==0.1.0 depends on anyio<=3, we can conclude that all versions of albatross depend on anyio<=3.
And because bird-feeder==1.0.0 depends on anyio>=4.3.0,<5 and only bird-feeder==1.0.0 is available, we can conclude that all versions of albatross and all versions of bird-feeder are incompatible.
And because albatross depends on albatross and bird-feeder, we can conclude that the requirements are unsatisfiable.
```
(Notice "albatross depends on albatross".)
## Summary
In a workspace, we now read configuration from the workspace root.
Previously, we read configuration from the first `pyproject.toml` or
`uv.toml` file in path -- but in a workspace, that would often be the
_project_ rather than the workspace configuration.
We need to read configuration from the workspace root, rather than its
members, because we lock the workspace globally, so all configuration
applies to the workspace globally.
As part of this change, the `uv-workspace` crate has been renamed to
`uv-settings` and its purpose has been narrowed significantly (it no
longer discovers a workspace; instead, it just reads the settings from a
directory).
If a user has a `uv.toml` in their directory or in a parent directory
but is _not_ in a workspace, we will still respect that use-case as
before.
Closes#4249.
## Summary
This PR introduces top-level configuration for uv, such that you can do:
```toml
[tool.uv]
index-url = "https://test.pypi.org/simple"
```
And `uv pip compile`, `uv run`, `uv tool run`, etc., will all respect
that configuration.
The settings that were escalated to the top-level remain on
`tool.uv.pip` too, but they're only respected in `uv pip` commands. If
they're specified in both places, then the `pip` settings win out.
While making this change, I also wired up some of the global options,
like `connectivity` and `native_tls`, through to all the relevant
places.
Closes#4250.
Previously, we took the first executable on the `PATH` but if it was not
a usable interpreter we'd fail. Now, we'll continue searching in the
path until we find an interpreter as we do with the standard executable
names.
Previously, `b` in the test case would have been incorrectly locked to
the path of `a`. I've moved `relative_to` into uv-fs since it's now used
in two different places.
Previously failing lockfile when `a/pyproject.toml` and
`a/b/pyproject.toml` exist (not in a workspace) and `a` was depending on
`b`:
```toml
version = 1
requires-python = ">=3.11, <3.13"
[[distribution]]
name = "b"
version = "0.1.0"
source = "directory+/home/konsti/projects/uv/a"
sdist = { path = "/home/konsti/projects/uv/a" }
[[distribution]]
name = "black"
version = "0.1.0"
source = "editable+."
sdist = { path = "." }
[[distribution.dependencies]]
name = "b"
version = "0.1.0"
source = "directory+/home/konsti/projects/uv/a"
```
Includes system interpreters in `uv toolchain list`.
This includes a refactor of `find_toolchain` to support iterating over
all toolchains
that match a request rather than ending earlier.
## Summary
This i still a draft, but it gets some of the work started on issue
#4064 , which requests --no-binary functionality similar to `uv pip
install` to exist on `uv pip compile`.
So far this has moved the command line shape from install and cloned it
over to compile. The actual functionality of respecting --no-binary
<package> and generating the resulting line in requirements.txt I could
use a hand with.
My understanding is we want to create a requirements.in file such as:
```
yt
```
Then when running `cargo run -- pip compile --no-binary yt
requirements.in`
we want the file to have this line in it:
```
yt==4.3.1 --no-binary yt
```
## Test Plan
Existing unit tests continue to pass.
No new unit tests have been created yet.
The new command line options do show up when testing with `cargo run --
pip compile -h`
---------
Co-authored-by: Charlie Marsh <charlie.r.marsh@gmail.com>
## Summary
Closes https://github.com/astral-sh/uv/issues/4296.
## Test Plan
Ran `cargo run lock --verbose` from
`scripts/workspaces/albatross-virtual-workspace`:
```
DEBUG uv 0.2.11 (ef3bc1612 2024-06-12)
warning: `uv lock` is experimental and may change without warning.
DEBUG Found workspace root: `/Users/crmarsh/workspace/puffin/scripts/workspaces/albatross-virtual-workspace`
DEBUG Adding discovered workspace member: /Users/crmarsh/workspace/puffin/scripts/workspaces/albatross-virtual-workspace/packages/albatross
DEBUG Adding discovered workspace member: /Users/crmarsh/workspace/puffin/scripts/workspaces/albatross-virtual-workspace/packages/bird-feeder
DEBUG Adding discovered workspace member: /Users/crmarsh/workspace/puffin/scripts/workspaces/albatross-virtual-workspace/packages/seeds
DEBUG Searching for Python >=3.12 in search path or managed toolchains
DEBUG Searching for managed toolchains at `/Users/crmarsh/Library/Application Support/uv/toolchains`
DEBUG Found managed toolchain `cpython-3.12.3-macos-aarch64-none`
DEBUG Found CPython 3.12.3 at `/Users/crmarsh/Library/Application Support/uv/toolchains/cpython-3.12.3-macos-aarch64-none/install/bin/python3` (managed toolchains)
Using Python 3.12.3 interpreter at: /Users/crmarsh/Library/Application Support/uv/toolchains/cpython-3.12.3-macos-aarch64-none/install/bin/python3
```
Before 0.2.10 we would parse `--python=python` as an executable name.
After https://github.com/astral-sh/uv/pull/4214, we started treating
this as a Python version range request (with an empty version range).
This is not entirely unreasonable, but it was an unexpected regression
and I don't think `VersionRequest` should support empty ranges in its
`from_str` implementation without more consideration.
Updates `--no-binary <package>` to take precedence over `--only-binary
:all:` and `--only-binary <package>` to take precedence over
`--no-binary :all:`.
I'm not entirely sure about this behavior, e.g. maybe I provided
`--only-binary :all:` later on the command line and really want it to
override those earlier arguments of `--no-binary <package>` for safety.
Right now we just fail to solve though since we can't satisfy the
overlapping requests.
Closes https://github.com/astral-sh/uv/issues/4063
## Summary
This is what I consider to be the "real" fix for #8072. We now treat
directory and path URLs as separate `ParsedUrl` types and
`RequirementSource` types. This removes a lot of `.is_dir()` forking
within the `ParsedUrl::Path` arms and makes some states impossible
(e.g., you can't have a `.whl` path that is editable). It _also_ fixes
the `direct_url.json` for direct URLs that refer to files. Previously,
we wrote out to these as if they were installed as directories, which is
just wrong.
We weren't using the common interface in `uv lock` because it didn't
support finding an interpreter without touching the virtual environment.
Here I refactor the project interface to support what we need and update
`uv lock` to use the shared implementation.
In the time before universal resolving, we would always pass a
`MarkerEnvironment`, and this environment would capture any relevant
`Requires-Python` specifier (including if `-p/--python` was provided on
the CLI).
But in universal resolution, we very specifically do not use a
`MarkerEnvironment` because we want to produce a resolution that is
compatible across potentially multiple environments. This in turn meant
that we lost `Requires-Python` filtering.
This PR adds it back. We do this by converting our `PythonRequirement`
into a `MarkerTree` that encodes the version specifiers in a
`Requires-Python` specifier. We then ask whether that `MarkerTree` is
disjoint with a dependency specification's `MarkerTree`. If it is, then
we know it's impossible for that dependency specification to every be
true, and we can completely ignore it.
## Summary
Right now, we're _always_ reinstalling local wheel archives, even if the
timestamp didn't change.
I want to fix the TODO properly but I will do so in a separate PR.
## Summary
`normalize` now takes an owned value and returns `Option<MarkerTree>`,
such that if any sub-expression evaluates to `true`, we can normalize
out the entire marker.
Closes https://github.com/astral-sh/uv/issues/4267.
Closes https://github.com/astral-sh/uv/issues/3857
Instead of using custom `Arch`, `Os`, and `Libc` types I just use
`target-lexicon`'s which enumerate way more variants and implement
display and parsing. We use a wrapper type to represent a couple special
cases to support the "x86" alias for "i686" and "macos" for "darwin".
Alternatively we could try to use our `platform-tags` types but those
capture more information (like operating system versions) that we don't
have for downloads.
As discussed in https://github.com/astral-sh/uv/pull/4160, this is not
sufficient for proper libc detection but that work is larger and will be
handled separately.
## Summary
Fix the docsting where `remove` was used instead of `add` in the context
of `uv add` command.
## Test Plan
```
cargo run -- add --help
```
```
Add one or more packages to the project requirements
Usage: uv add [OPTIONS] <REQUIREMENTS>...
Arguments:
<REQUIREMENTS>...
The packages to add, as PEP 508 requirements (e.g., `flask==2.2.3`)
```
## Summary
I think this is a useful piece of connective tissue that will let us
avoid back-and-forths when folks include traces.
## Test Plan
```
❯ cargo run pip list --verbose
DEBUG uv 0.2.11 (44041bccd 2024-06-11)
DEBUG Searching for Python interpreter in virtual environments
DEBUG Found CPython 3.12.3 at `/Users/crmarsh/workspace/puffin/.venv/bin/python3` (virtual environment)
DEBUG Using Python 3.12.3 environment at .venv/bin/python3
```
Zanie Blue
Chan Kang
Ibraheem Ahmed
konsti
Andrew Gallant
Michał Górny
Charlie Marsh
samypr100
Brian Mego
Nicolas Delaby