## Summary
The issue here is fairly complex. Consider the following:
```toml
[project]
name = "project"
version = "0.1.0"
requires-python = ">=3.12.0"
dependencies = []
[project.optional-dependencies]
cpu = [
"torch>=2.5.1",
"torchvision>=0.20.1",
]
cu124 = [
"torch>=2.5.1",
"torchvision>=0.20.1",
]
[tool.uv]
conflicts = [
[
{ extra = "cpu" },
{ extra = "cu124" },
],
]
[tool.uv.sources]
torch = [
{ index = "pytorch-cpu", extra = "cpu", marker = "platform_system != 'Darwin'" },
]
torchvision = [
{ index = "pytorch-cpu", extra = "cpu", marker = "platform_system != 'Darwin'" },
]
[[tool.uv.index]]
name = "pytorch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true
```
When solving this project, we first pick a PyTorch version from PyPI, to
solve the `cu124` extra, selecting `2.5.1`.
Later, we try to solve the `cpu` extra. In solving that extra, we look
at the PyTorch CPU index. Ideally, we'd select `2.5.1+cpu`... But
`2.5.1` is already a preference. So we choose that.
Now, we only respect preferences for explicit indexes if they came from
the same index.
Closes https://github.com/astral-sh/uv/issues/9295.
## Summary
This PR enables something like the "final boss" of PyTorch setups --
explicit support for CPU vs. GPU-enabled variants via extras:
```toml
[project]
name = "project"
version = "0.1.0"
requires-python = ">=3.13.0"
dependencies = []
[project.optional-dependencies]
cpu = [
"torch==2.5.1+cpu",
]
gpu = [
"torch==2.5.1",
]
[tool.uv.sources]
torch = [
{ index = "torch-cpu", extra = "cpu" },
{ index = "torch-gpu", extra = "gpu" },
]
[[tool.uv.index]]
name = "torch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true
[[tool.uv.index]]
name = "torch-gpu"
url = "https://download.pytorch.org/whl/cu124"
explicit = true
[tool.uv]
conflicts = [
[
{ extra = "cpu" },
{ extra = "gpu" },
],
]
```
It builds atop the conflicting extras work to allow sources to be marked
as specific to a dedicated extra being enabled or disabled.
As part of this work, sources now have an `extra` field. If a source has
an `extra`, it means that the source is only applied to the requirement
when defined within that optional group. For example, `{ index =
"torch-cpu", extra = "cpu" }` above only applies to
`"torch==2.5.1+cpu"`.
The `extra` field does _not_ mean that the source is "enabled" when the
extra is activated. For example, this wouldn't work:
```toml
[project]
name = "project"
version = "0.1.0"
requires-python = ">=3.13.0"
dependencies = ["torch"]
[tool.uv.sources]
torch = [
{ index = "torch-cpu", extra = "cpu" },
{ index = "torch-gpu", extra = "gpu" },
]
[[tool.uv.index]]
name = "torch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true
[[tool.uv.index]]
name = "torch-gpu"
url = "https://download.pytorch.org/whl/cu124"
explicit = true
```
In this case, the sources would effectively be ignored. Extras are
really confusing... but I think this is correct? We don't want enabling
or disabling extras to affect resolution information that's _outside_ of
the relevant optional group.
## Summary
I was wrongly using `.name()` to detect if a package was "not root", but
in `pip compile`, the root can have a name -- so we were failing to find
the derivation chain.
## Summary
This PR adds context to our error messages to explain _why_ a given
package was included, if we fail to download or build it.
It's quite a large change, but it motivated some good refactors and
improvements along the way.
Closes https://github.com/astral-sh/uv/issues/8962.
## Summary
As discussed in Discord... This struct has evolved to include a lot of
information apart from the `petgraph::Graph`. And I want to add a graph
to the simplified `Resolution` type. So I think this name makes more
sense.
This adds support for providing conflicting group names in addition to
extra names to `Conflicts`.
This merely makes "room" for it in the types while keeping everything
working. We'll add proper support for it in the next commit.
Note that one interesting trick we do here is depend directly on
`hashbrown` so that we can make use of its `Equivalent` trait. This in
turn lets us use things like `ConflictItemRef` as a lookup key for a
hashset that contains `ConflictItem`. This mirrors using a `&str` as a
lookup key for a hashset that contains `String`, but works for arbitrary
types. `std` doesn't support this, but `hashbrown` does. This trick in
turn lets us simplify some of our data structures.
This also rejiggers some of the serde-interaction with the conflicting
types. We now use a wire type to represent our conflicting items for
more flexibility. i.e., Support `extra` XOR `group` fields.
Since this is intended to support _both_ groups and extras, it doesn't
make sense to just name it for groups. And since there isn't really a
word that encapsulates both "extra" and "group," we just fall back to
the super general "conflicts."
We'll rename the variables and other things in the next commit.
## Summary
I need this for the derivation chain work
(https://github.com/astral-sh/uv/issues/8962), but it just seems
generally useful. You can't always get a version from a `Dist` (it could
be URL-based!), but when we create a `ResolvedDist`, we _do_ know the
version (and not just the URL). This PR preserves it.
This PR adds support for conflicting extras. For example, consider
some optional dependencies like this:
```toml
[project.optional-dependencies]
project1 = ["numpy==1.26.3"]
project2 = ["numpy==1.26.4"]
```
These dependency specifications are not compatible with one another.
And if you ask uv to lock these, you'll get an unresolvable error.
With this PR, you can now add this to your `pyproject.toml` to get
around this:
```toml
[tool.uv]
conflicting-groups = [
[
{ package = "project", extra = "project1" },
{ package = "project", extra = "project2" },
],
]
```
This will make the universal resolver create additional forks
internally that keep the dependencies from the `project1` and
`project2` extras separate. And we make all of this work by reporting
an error at **install** time if one tries to install with two or more
extras that have been declared as conflicting. (If we didn't do this,
it would be possible to try and install two different versions of the
same package into the same environment.)
This PR does *not* add support for conflicting **groups**, but it is
intended to add support in a follow-up PR.
Closes#6981Fixes#8024
Ref #6729, Ref #6830
This should also hopefully unblock
https://github.com/dagster-io/dagster/pull/23814, but in my testing, I
did run into other problems (specifically, with `pywin`). But it does
resolve the problem with incompatible dependencies in two different
extras once you declare `test-airflow-1` and `test-airflow-2` as
conflicting for `dagster-airflow`.
NOTE: This PR doesn't make `conflicting-groups` public yet. And in a
follow-up PR, I plan to switch the name to `conflicts` instead of
`conflicting-groups`, since it will be able to accept conflicting extras
_and_ conflicting groups.
## Summary
We're inconsistent with these -- sometimes it's `Error::Fetch` and
sometimes it's `Error::Download`. The message says download, so let's
just use that?
After https://github.com/astral-sh/uv/pull/8797, we have spec-compliant
handling for local version identifiers and can completely remove all the
special-casing around it.
Implement a full working version of local version semantics. The (AFAIA)
major move towards this was implemented in #2430. This added support
such that the version specifier `torch==2.1.0+cpu` would install
`torch@2.1.0+cpu` and consider `torch@2.1.0+cpu` a valid way to satisfy
the requirement `torch==2.1.0` in further dependency resolution.
In this feature, we more fully support local version semantics. Namely,
we now allow `torch==2.1.0` to install `torch@2.1.0+cpu` regardless of
whether `torch@2.1.0` (no local tag) actually exists.
We do this by adding an internal-only `Max` value to local versions that
compare greater to all other local versions. Then we can translate
`torch==2.1.0` into bounds: greater than 2.1.0 with no local tag and
less than 2.1.0 with the `Max` local tag.
Depends on https://github.com/astral-sh/packse/pull/227.
This updates the surrounding code to use the new ResolverEnvironment
type. In some cases, this simplifies caller code by removing case
analysis. There *shouldn't* be any behavior changes here. Some test
snapshots were updated to account for some minor tweaks to error
messages.
I didn't split this up into separate commits because it would have been
too difficult/costly.
This type is intended to replace `ResolverMarkers`. The main difference
between them is that this type encapsulates more decision making by
un-exporting the different cases. So instead of callers needing to do
explicit case analysis depending on the type of resolver environment,
callers instead use methods that know how to do the right thing. In the
next commit, there are at least a few cases where this greatly
simplifies case analysis on the caller side.
The motivation for this type is to centralize decision making about
forking. In particular, we want to expand forking to include conflicting
groups instead of just `MarkerTree`. So to a certain extent, the
refactor here is about removing bare use of `MarkerTree` in favor of a
more purpose built type that encapsulates the forking logic.
The encapsulation is not quite perfect here. I expect to improve on it a
bit once we add support for conflicting groups.
This is split off from the subsequent commit (that makes use of
`ResolverEnvironment`) so that it's a bit easier to review the addition
in isolation.
## Summary
At present, when we have a Python requirement and we see a wheel, we
verify that the Python requirement is compatible with the wheel. For
source distributions, though, we verify that both the Python requirement
_and_ the currently-installed version are compatible, because we assume
that we'll need to build the source distribution in order to get
metadata. However, we can often extract source distribution metadata
_without_ building (e.g., if there's a `pyproject.toml` with no dynamic
keys).
This PR thus modifies the source distribution handling to defer that
incompatibility ("We couldn't get metadata for this project, because it
has no static metadata and requires a higher Python version to run /
build") until we actually try to build the package. As a result, you can
now resolve source distribution-only packages using Python versions
below their `requires-python`, as long as they include static metadata.
Closes https://github.com/astral-sh/uv/issues/8767.
This PR fixes a bug where it was possible for dependencies to be
included in a final resolution with markers that always evaluate to
false. Specifically, `python_version < '0'`.
While we do filter based on Python markers during forking, it turns out
that the markers for each fork are "combined" *after* this filtering
step. But the process of combination can result in a more specific
marker that is always false for the configured Python requirement. This
could result in dependencies with markers that are always false (like
`python_version < '0'`) appearing in the resolution.
The first commit in this PR adds a regression test (with an undesirable
result), and the second commit fixes the regression and updates the
test.
Fixes#8676
Previously, when doing a `uv pip` resolution, we would only return the
first entry in the map. But there should only ever be one entry, or else
we would have incompatible dependencies. So we can collapse the case
with the "one universal fork" case.
(I found this while doing some refactoring of how we handle forking, and
collapsing these cases simplifies some of that refactoring work.)
## Summary
If you pass a named index via the CLI, you can now reference it as a
named source. This required some surprisingly large refactors, since we
now need to be able to track whether a given index was provided on the
CLI vs. elsewhere (since, e.g., we don't want users to be able to
reference named indexes defined in global configuration).
Closes https://github.com/astral-sh/uv/issues/7899.
## Summary
This PR lifts the restriction that a package must come from a single
index. For example, you can now do:
```toml
[project]
name = "project"
version = "0.1.0"
readme = "README.md"
requires-python = ">=3.12"
dependencies = ["jinja2"]
[tool.uv.sources]
jinja2 = [
{ index = "torch-cu118", marker = "sys_platform == 'darwin'"},
{ index = "torch-cu124", marker = "sys_platform != 'darwin'"},
]
[[tool.uv.index]]
name = "torch-cu118"
url = "https://download.pytorch.org/whl/cu118"
[[tool.uv.index]]
name = "torch-cu124"
url = "https://download.pytorch.org/whl/cu124"
```
The construction is very similar to the way we handle URLs today: you
can have multiple URLs for a given package, but they must appear in
disjoint forks. So most of the code is just adding that abstraction to
the resolver, following our handling of URLs.
Closes#7761.
## Summary
This PR adds a first-class API for defining registry indexes, beyond our
existing `--index-url` and `--extra-index-url` setup.
Specifically, you now define indexes like so in a `uv.toml` or
`pyproject.toml` file:
```toml
[[tool.uv.index]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu121"
```
You can also provide indexes via `--index` and `UV_INDEX`, and override
the default index with `--default-index` and `UV_DEFAULT_INDEX`.
### Index priority
Indexes are prioritized in the order in which they're defined, such that
the first-defined index has highest priority.
Indexes are also inherited from parent configuration (e.g., the
user-level `uv.toml`), but are placed after any indexes in the current
project, matching our semantics for other array-based configuration
values.
You can mix `--index` and `--default-index` with the legacy
`--index-url` and `--extra-index-url` settings; the latter two are
merely treated as unnamed `[[tool.uv.index]]` entries.
### Index pinning
If an index includes a name (which is optional), it can then be
referenced via `tool.uv.sources`:
```toml
[[tool.uv.index]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu121"
[tool.uv.sources]
torch = { index = "pytorch" }
```
If an index is marked as `explicit = true`, it can _only_ be used via
such references, and will never be searched implicitly:
```toml
[[tool.uv.index]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu121"
explicit = true
[tool.uv.sources]
torch = { index = "pytorch" }
```
Indexes defined outside of the current project (e.g., in the user-level
`uv.toml`) can _not_ be explicitly selected.
(As of now, we only support using a single index for a given
`tool.uv.sources` definition.)
### Default index
By default, we include PyPI as the default index. This remains true even
if the user defines a `[[tool.uv.index]]` -- PyPI is still used as a
fallback. You can mark an index as `default = true` to (1) disable the
use of PyPI, and (2) bump it to the bottom of the prioritized list, such
that it's used only if a package does not exist on a prior index:
```toml
[[tool.uv.index]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu121"
default = true
```
### Name reuse
If a name is reused, the higher-priority index with that name is used,
while the lower-priority indexes are ignored entirely.
For example, given:
```toml
[[tool.uv.index]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu121"
[[tool.uv.index]]
name = "pytorch"
url = "https://test.pypi.org/simple"
```
The `https://test.pypi.org/simple` index would be ignored entirely,
since it's lower-priority than `https://download.pytorch.org/whl/cu121`
but shares the same name.
Closes#171.
## Future work
- Users should be able to provide authentication for named indexes via
environment variables.
- `uv add` should automatically write `--index` entries to the
`pyproject.toml` file.
- Users should be able to provide multiple indexes for a given package,
stratified by platform:
```toml
[tool.uv.sources]
torch = [
{ index = "cpu", markers = "sys_platform == 'darwin'" },
{ index = "gpu", markers = "sys_platform != 'darwin'" },
]
```
- Users should be able to specify a proxy URL for a given index, to
avoid writing user-specific URLs to a lockfile:
```toml
[[tool.uv.index]]
name = "test"
url = "https://private.org/simple"
proxy = "http://<omitted>/pypi/simple"
```
As per
https://matklad.github.io/2021/02/27/delete-cargo-integration-tests.html
Before that, there were 91 separate integration tests binary.
(As discussed on Discord — I've done the `uv` crate, there's still a few
more commits coming before this is mergeable, and I want to see how it
performs in CI and locally).
#7226 modified the check to skip prefetching of source dists without
proper minimum-version bounds, and wound up flipping the boolean
expression. This change flips the some/none expression so that the
intended skip happens as expected.
Fixes#7680.
This is preparatory work for the upload functionality, which needs to
read the METADATA file and attach its parsed contents to the POST
request: We move finding the `.dist-info` from `install-wheel-rs` and
`uv-client` to a new `uv-metadata` crate, so it can be shared with the
publish crate.
I don't properly know if its the right place since the upload code isn't
ready, but i'm PR-ing it now because it already had merge conflicts.
## Summary
We now track the discovered `IndexCapabilities` for each `IndexUrl`. If
we learn that an index doesn't support range requests, we avoid doing
any batch prefetching.
Closes https://github.com/astral-sh/uv/issues/7221.
## Summary
If we have a singleton `Range`, we don't need to iterate over the map of
available ranges; instead, we can just get the singleton directly.
Closes#6131.
## Summary
I think a better tradeoff here is to skip fetching metadata, even though
we can't validate the extras.
It will help with situations like
https://github.com/astral-sh/uv/issues/5073#issuecomment-2334235588 in
which, otherwise, we have to download the wheels twice.
This commit refactors how deal with `requires-python` so that instead of
simplifying markers of dependencies inside the resolver, we do it at the
edges of our system. When writing markers to output, we simplify when
there's an obvious `requires-python` context. And when reading markers
as input, we complexity markers with the relevant `requires-python`
constraint.
When I first wrote this routine, it was intended to only emit a trace
for the final "unioned" resolution. But we actually moved that semantic
operation to the construction of the resolution *graph*. So there is no
unioned `Resolution` any more.
But this is still useful to see. So I changed this to just emit a trace
of *every* resolution right before constructing the graph.
It might be nice to also emit a trace of the unioned graph too. Or
perhaps we should do that instead if this proves too noisy. (Although
this is only emitted at TRACE level.)
## Summary
Right now, we have slightly different `requires-python` semantics for
`-p 3.11` vs. `-p 3.11 --universal`, and slightly different (wrong)
semantics for how we compare against the _installed_ Python version
(which doesn't ignore upper bounds, but should).
This PR rips it all out and replaces it with consistent semantics across
`uv lock`, `uv pip compile -p 3.11`, and `uv pip compile -p 3.11
--universal`. We now always ignore upper bounds.
Closes https://github.com/astral-sh/uv/issues/6859.
Closes https://github.com/astral-sh/uv/issues/5045.
## Summary
We now respect the user-provided upper-bound in for `requires-python`.
So, if the user has `requires-python = "==3.11.*"`, we won't explore
forks that have `python_version >= '3.12'`, for example.
However, we continue to _only_ compare the lower bounds when assessing
whether a dependency is compatible with a given Python range.
Closes https://github.com/astral-sh/uv/issues/6150.
## Summary
This is similar to https://github.com/astral-sh/uv/pull/6171 but more
expansive... _Anywhere_ that we test requirements for platform
compatibility, we _need_ to respect the resolver-friendly markers. In
fixing the motivating issue (#6621), I also realized that we had a bunch
of bugs here around `pip install` with `--python-platform` and
`--python-version`, because we always performed our `satisfy` and `Plan`
operations on the interpreter's markers, not the adjusted markers!
Closes https://github.com/astral-sh/uv/issues/6621.
Closes https://github.com/astral-sh/uv/issues/6167
We've been seeing intermittent failures in CI, which we thought were
unexpected HTTP 401s but it actually looks like a panic when handling an
expected HTTP error. I believe the problem is that an early client error
can cause the channel to close and we crash when we unwrap the `send`.
## Summary
In the resolver, we use release-only semantics to normalize
`python_full_version`. So, if we see `python_full_version < '3.13'`, we
treat that as `(Unbounded, Exclude(3.13))`. `3.13b0` evaluates as `true`
to that range, so we were accepting pre-releases for these markers.
Instead, we need to exclude pre-release segments when performing these
evaluations.
Closes https://github.com/astral-sh/uv/issues/6169.
## Test Plan
Hard to write a test for this because you need a pre-release Python
locally... so:
`echo "sqlalchemy==2.0.32" | cargo run pip compile - --python 3.13 -n`
There were different `PubGrubPackage` types so they never matched the
available versions set! Luckily, the available versions are agnostic to
the markers and optional dependencies so we can just broaden to using
`PackageName` as a lookup key.
Addresses yet another complaint in
https://github.com/astral-sh/uv/issues/5046
We have bad error messages for optional (extra) dependencies and
development dependencies in workspaces:
1. We weren't showing the full package, so we'd drop `:dev` and
`[extra]` by accident
2. We didn't include derived packages, e.g., `member[extra]` in tree
processing collapse operation, so we'd include extra clauses like the
ones we removed in #6092
Also
- Reverts
f0de4f71f2
— it turns out it wasn't quite correct and it didn't seem worth using
the custom incompatibility anymore.
- Fixes a bug in the display of `package:dev` which was not showing
`:dev` for some variants (see 94d8020b58)
## Summary
Normalize all `python_version` markers to their equivalent
`python_full_version` form. This avoids false positives in forking
because we currently cannot detect any relationships between the two
forms. It also avoids subtle bugs due to the truncating semantics of
`python_version`. For example, given `requires-python = ">3.12"`, we
currently simplify the marker `python_version <= 3.12` to `false`.
However, the version `3.12.1` will be truncated to `3.12` for
`python_version` comparisons, and thus it satisfies the python
requirement and evaluates to `true`.
It is possible to simplify back to `python_version` when writing markers
to the lockfile. However, the equivalent `python_full_version` markers
are often clearer and easier to simplify, so I lean towards leaving them
as `python_full_version`.
There are *a lot* of snapshot updates from this change. I'd like more
eyes on the transformation logic in `python_version_to_full_version` to
ensure that they are all correct.
Resolves https://github.com/astral-sh/uv/issues/6125.
## Summary
Using https://github.com/astral-sh/uv/issues/6064 as a motivating
example: at present, on main, we're not properly propagating the
`Requires-Python` simplifications. In that case, for example, we end up
solving for a branch with `python_version < 3.11`, and a branch `>=
3.11`, even though `Requires-Python` is `>=3.11`. Later, when we get to
the graph, we apply version simplification based on `Requires-Python`,
which causes us to _remove_ the `python_version < 3.11` markers
entirely, leaving us with duplicate dependencies for `pylint`.
This PR instead tries to ensure that we always apply this narrowing to
requirements and forks, so that we don't need to apply the same
simplification when constructing the graph at all.
Closes https://github.com/astral-sh/uv/issues/6064.
Closes#6059.
In particular, I added this as a hack to avoid a kinda of
instability that was caused by our marker code not correctly
detecting markers that were always false. But that has since
been fixed.
Removing this code doesn't change any tests. Arguably it
should be possible to come up with a test that failed with
this hack inserted but succeeded without it. In particular,
with this hack, new forks were being prevented from being
added even when they ought to be added, e.g., when preferences
get updated.
An extension of #6090 that replaces #6066.
In brief,
1. Workspace member names are passed to the resolver for no solution
errors
2. There is a new derivation tree pre-processing step that trims
`NoVersion` incompatibilities for workspace members from the derivation
tree. This avoids showing redundant clauses like `Because only
bird==0.1.0 is available and bird==0.1.0 depends on anyio==4.3.0, we can
conclude that all versions of bird depend on anyio==4.3.0.`. As a minor
note, we use a custom incompatibility kind to mark these
incompatibilities at resolution-time instead of afterwards.
3. Root dependencies on workspace members say `your workspace requires
bird` rather than `you require bird`
4. Workspace member package display omits the version, e.g., `bird`
instead of `bird==0.1.0`
5. Instead of reporting a workspace member as unusable we note that its
requirements cannot be solved, e.g., `bird's requirements are
unsatisfiable` instead of `bird cannot be used`.
6. Instead of saying `your requirements are unsatisfiable` we say `your
workspace's requirements are unsatisfiable` when in a workspace, since
we're not in a "provide direct requirements" paradigm.
As an annoying but minor implementation detail, `PackageRange` now
requires access to the `PubGrubReportFormatter` so it can determine if
it is formatting a workspace member or not. We could probably improve
the abstractions in the future.
As a follow-up, we should additional special casing for "single project"
workspaces to avoid mention of the workspace concept in simple projects.
However, it looks like this will require additional tree manipulations
so I'm going to keep it separate.
## Summary
Historically, in order to "resolve from a lockfile", we've taken the
lockfile, used it to pre-populate the in-memory metadata index, then run
a resolution. If the resolution didn't match our existing resolution, we
re-resolved from scratch.
This was an appealing approach because (in theory) it didn't require any
dedicated logic beyond pre-populating the index. However, it's proven to
be _really_ hard to get right, because it's a stricter requirement than
we need. We just need the current lockfile to _satisfy_ the requirements
provided by the user. We don't actually need a second resolution to
produce the exact same result. And it's not uncommon that this second
resolution differs, because we seed it with preferences, which
fundamentally changes its course. We've worked hard to minimize those
"instabilities", but they're still present.
The approach here is intended to be much simpler. Instead of resolving
from the lockfile, we just check if the current resolution satisfies the
state of the workspace. Specifically, we check if the lockfile (1)
contains all the relevant members, and (2) matches the metadata for all
dependencies, recursively. (We skip registry dependencies, assuming that
they're immutable.)
This may actually be too conservative, since we can have resolutions
that satisfy the requirements, even if the requirements have changed
slightly. But we want to bias towards correctness for now.
My hope is that this scheme will be more performant, simpler, and more
robust.
Closes https://github.com/astral-sh/uv/issues/6063.
## Summary
This was added in https://github.com/astral-sh/uv/pull/5405 but is now
the cause of an instability in `github_wikidata_bot`. Specifically, on
the initial run, we fork in `pydantic==2.8.2`, via:
```
Requires-Dist: typing-extensions>=4.12.2; python_version >= '3.13'
Requires-Dist: typing-extensions>=4.6.1; python_version < '3.13'
```
In the end, we resolve a single version of `typing-extensions`
(`4.12.2`)... But we don't recognize the two resolutions as the "same
graph", because we propagate the fork markers, and so the "edges" have
different markers on them...
In the second run through, when we have the forks in advance, we don't
split on Pydantic... We just try to solve from the root with the current
forks. This is fundamentally different and I fear it will be the cause
of many instabilities. But removing this graph check fixes the proximate
issue.
I don't really understand why this was added since there was no test
coverage in the PR.
## Summary
When constructing the `Resolution`, we only propagated the fork markers
to the package node, but not the extras node. This led to cases in which
an extra could be included unconditionally or otherwise diverge from the
base package version.
Closes https://github.com/astral-sh/uv/issues/6062.
## Summary
Right now, we store the environment markers in a `BTreeSet` -- so
they're sorted, but the sort doesn't really tell us anything. I think we
should instead store them in the order in which we solved. I thought
this might fix an instability (it didn't), but I think it's still good
to ensure we solve in the same order.
I also changed from `Option<Vec>` to just `Vec`, since there was no
distinction between `None` and empty.
Warn when there are missing bounds on transitive dependencies with
`--resolution lowest`.
Implemented as a lazy resolution graph check. Dev deps are odd because
they are missing the edge from the root that extras have (they are
currently orphans in the resolution graph), but this is more complex to
solve properly because we can put dev dep information in a `Requirement`
so i special cased them here.
Closes#2797
Should help with #1718
---------
Co-authored-by: Ibraheem Ahmed <ibraheem@ibraheem.ca>
## Summary
This PR rewrites the `MarkerTree` type to use algebraic decision
diagrams (ADD). This has many benefits:
- The diagram is canonical for a given marker function. It is impossible
to create two functionally equivalent marker trees that don't refer to
the same underlying ADD. This also means that any trivially true or
unsatisfiable markers are represented by the same constants.
- The diagram can handle complex operations (conjunction/disjunction) in
polynomial time, as well as constant-time negation.
- The diagram can be converted to a simplified DNF form for user-facing
output.
The new representation gives us a lot more confidence in our marker
operations and simplification, which is proving to be very important
(see https://github.com/astral-sh/uv/pull/5733 and
https://github.com/astral-sh/uv/pull/5163).
Unfortunately, it is not easy to split this PR into multiple commits
because it is a large rewrite of the `marker` module. I'd suggest
reading through the `marker/algebra.rs`, `marker/simplify.rs`, and
`marker/tree.rs` files for the new implementation, as well as the
updated snapshots to verify how the new simplification rules work in
practice. However, a few other things were changed:
- [We now use release-only comparisons for `python_full_version`, where
we previously only did for
`python_version`](https://github.com/astral-sh/uv/blob/ibraheem/canonical-markers/crates/pep508-rs/src/marker/algebra.rs#L522).
I'm unsure how marker operations should work in the presence of
pre-release versions if we decide that this is incorrect.
- [Meaningless marker expressions are now
ignored](https://github.com/astral-sh/uv/blob/ibraheem/canonical-markers/crates/pep508-rs/src/marker/parse.rs#L502).
This means that a marker such as `'x' == 'x'` will always evaluate to
`true` (as if the expression did not exist), whereas we previously
treated this as always `false`. It's negation however, remains `false`.
- [Unsatisfiable markers are written as `python_version <
'0'`](https://github.com/astral-sh/uv/blob/ibraheem/canonical-markers/crates/pep508-rs/src/marker/tree.rs#L1329).
- The `PubGrubSpecifier` type has been moved to the new `uv-pubgrub`
crate, shared by `pep508-rs` and `uv-resolver`. `pep508-rs` also depends
on the `pubgrub` crate for the `Range` type, we probably want to move
`pubgrub::Range` into a separate crate to break this, but I don't think
that should block this PR (cc @konstin).
There is still some remaining work here that I decided to leave for now
for the sake of unblocking some of the related work on the resolver.
- We still use `Option<MarkerTree>` throughout uv, which is unnecessary
now that `MarkerTree::TRUE` is canonical.
- The `MarkerTree` type is now interned globally and can potentially
implement `Copy`. However, it's unclear if we want to add more
information to marker trees that would make it `!Copy`. For example, we
may wish to attach extra and requires-python environment information to
avoid simplifying after construction.
- We don't currently combine `python_full_version` and `python_version`
markers.
- I also have not spent too much time investigating performance and
there is probably some low-hanging fruit. Many of the test cases I did
run actually saw large performance improvements due to the markers being
simplified internally, reducing the stress on the old `normalize`
routine, especially for the extremely large markers seen in
`transformers` and other projects.
Resolves https://github.com/astral-sh/uv/issues/5660,
https://github.com/astral-sh/uv/issues/5179.
## Summary
This _used_ to be true but we now require fetching metadata for all
distributions even with `--no-deps` since, e.g., we validate that any
declared extras exist.
## Summary
This fixes a bug introduced by
https://github.com/astral-sh/uv/pull/5232. It turns out that the
`universal_disjoint_base_or_local_requirement` test does not actually do
what it was meant to because of the incorrect python requirement. With a
valid python requirement, it fails on `main`. The problem is that we try
to exclude the original base version from the range of allowed versions
to try and prefer local versions. However, in the test, there is a
branch that depends on the non-local version, with no applicable local
in its fork. We should remove this exclusion as prioritization is
handled by the candidate resolver.
## Summary
If _both_ nodes in the derivation tree are proxies, we need to remove
the _entire_ node. So, the function now returns an `Option<Tree>`
instead of taking `&mut Tree`.
Closes https://github.com/astral-sh/uv/issues/5618.
By resolving for each fork from the lockfile individually and by adding
using preferences for the current fork, we solve the instability #5180.
I've tested the locally and will add the packse test scenarios upstack.
Part of
https://github.com/astral-sh/uv/issues/5180#issuecomment-2247696198
## Summary
Given a fork like:
```
pylint < 3 ; sys_platform == 'darwin'
pylint > 2 ; sys_platform != 'darwin'
```
Solving the top branch will typically yield a solution that also
satisfies the bottom branch, due to maximum version selection (while the
inverse isn't true).
To quote an example from the docs:
```rust
// If there's no difference, prioritize forks with upper bounds. We'd prefer to solve
// `numpy <= 2` before solving `numpy >= 1`, since the resolution produced by the former
// might work for the latter, but the inverse is unlikely to be true due to maximum
// version selection. (Selecting `numpy==2.0.0` would satisfy both forks, but selecting
// the latest `numpy` would not.)
```
Closes https://github.com/astral-sh/uv/issues/4926 for now.
## Summary
First part of: https://github.com/astral-sh/uv/issues/4926. We should
solve forks that _don't_ expand the world of supported versions (e.g.,
`python_version >= '3.11'` enables us to select new packages, since we
narrow the supported version range).
This PR represents a different approach to marker propagation in an
attempt to unblock #4640. In particular, instead of propagating markers
when forks are created, we wait until resolution is complete to
propagate all markers to all dependencies in each fork. This ends up
being both more robust (we should never miss anything) and simpler to
implement because it doesn't require mutating a `PubGrubPackage` (which
was pretty annoying). I think the main downside here is that this can
sometimes add markers where they aren't needed.
This actually winds up making quite a few snapshot changes. I went
through each of them. Some of them look like legitimate bug fixes. Some
of them look like superfluous additions. And some of them look like they
would be removed if we had perfect marker normalization. But I don't
think any of the changes are _wrong_.
Consider the following packse scenario:
```toml
[root]
requires = [
"a>=1.0.0 ; python_version < '3.10'",
"a>=1.1.0 ; python_version >= '3.10'",
"a>=1.2.0 ; python_version >= '3.11'",
]
[packages.a.versions."1.0.0"]
[packages.a.versions."1.1.0"]
[packages.a.versions."1.2.0"]
```
On current `main`, this produces a dependency on `a` that looks like
this:
```toml
dependencies = [
{ name = "fork-overlapping-markers-basic-a", marker = "python_version < '3.10' or python_version >= '3.11'" },
]
```
But the marker expression is clearly wrong here, since it implies that
`a` isn't installed at all for Python 3.10. With this PR, the above
dependency becomes:
```toml
dependencies = [
{ name = "fork-overlapping-markers-basic-a" },
]
```
That is, it's unconditional. Which is I believe correct here since there
aren't any other constraints on which version to select.
The specific bug here is that when we found overlapping dependency
specifications for the same package *within* a pre-existing fork, we
intersected all of their marker expressions instead of unioning them.
That in turn resulted in incorrect marker expressions.
While this doesn't fix any known bug on the issue tracker (like #4640),
it does appear to fix a couple of our snapshot tests. And fixes a basic
test case I came up with while working on #4732.
For the packse scenario test: https://github.com/astral-sh/packse/pull/206
When a fork occurs, we divide not just the dependencies that
provoked a fork into distinct groups, but we also add the
corresponding sibling dependencies to each fork. Previously,
while we track markers on the fork itself, the individual
dependencies that had markers only corresponded to markers
written from the dependency specification.
This meant that the sibling dependencies that got added to
each fork would not themselves have markers attached to them.
This in turn meant they would not have markers associated with
them in the lock file.
In many cases, this is actually okay, because the resolver will
pick a version that is "universal" across all forks in most
cases. But in some cases, this just simply isn't possible as
the marker expressions in the fork can and do influence resolution.
In which case, it is possible for the same package with different
versions to show up in the lock file unconditionally. Which is a
big no-no.
So in this commit, after we determine the forks, we intersect the
markers on each fork with each of its dependencies.
This does seem to balloon the marker expressions in some cases.
I plucked one low hanging fruit to avoid doing `x and x` in
trivial cases. (And this eliminated a portion of the snapshot
diffs.) But some pretty gnarly diffs remain.
This commit also fixes another bug: previously, when we created a fork
to capture the "remaining" universe of an incomplete set of markers, we
left out dependencies that should be included in that fork. We rectify
that here.
Fixes#5086
Partially addresses #4732
Consider these requirements from pylint 3.2.5:
```
Requires-Dist: dill >=0.3.6 ; python_version >= "3.11"
Requires-Dist: dill >=0.3.7 ; python_version >= "3.12"
```
We will split on the python version, but then we may pick a version of
`dill` that's `>=0.3.7` in both branches and also have an otherwise
identical resolution in both forks. In this case, we merge both forks
and store only their conjoined markers.
## Summary
Prior to this change, the resolver would panic if we ran with
`--offline` and `--no-deps` and we had cached metadata for a _package_
(i.e., the versions) but no cached metadata for the _distribution_
(i.e., the specific wheel), since we weren't validating that the
returned metadata in the `--no-deps` case was actually successful. (We
need metadata, even for `--no-deps`, so that we can validate extras.)
## Test Plan
The added test panics on the previous branch.
## Summary
Excellent find from @konstin. If we have a package that's included in
two forks at the same version, but with different URLs, we need to avoid
collapsing them in the lockfile.
Closes https://github.com/astral-sh/uv/issues/5294.
Looking at how to merge identical forks, i found that the `Resolution`
can be simplified by treating it as a nodes and edges store (plus pins,
they are separate since they are per name-version, not per
(virtual-)package-version). This should also make #5294 more apparent,
which i didn't touch here.
I additionally added some doc comments to the `Resolution` types.
Our everything-method `solve` tends to grow large, so before i'm adding
more logic, i'm moving some code and some logging statements around to
keep it manageable.
I made minor changes to the logging, otherwise no logic changes, only
refactoring.
## Summary
This PR modifies the lockfile to include the impactful resolution
settings, like the resolution and pre-release mode. If any of those
values change, we want to ignore the existing lockfile. Otherwise,
`--resolution lowest-direct` will typically have no effect, which is
really unintuitive.
Closes https://github.com/astral-sh/uv/issues/5226.
Andrew Gallant
Charlie Marsh
Eric Mark Martin
konsti
Amos Wenger
Topher Anderson
Zanie Blue
Ibraheem Ahmed