Solving spent a chunk of its time just converting resolutions, the left
two blocks:
These blocks are `ResolverOutput::from_state` with 1.3% and
`ForkState::into_resolution` with 4.1% of resolver thread runtime for
apache airflow universal.
We reduce the overhead spent in those functions, to now 1.1% and 2.1% of
resolver time spend in those functions by:
Commit 1: Replace the hash set for the edges with a vec in
`ForkState::into_resolution`. We deduplicate edges anyway when
collecting them, and the hash-and-insert was slow.
Commit 2: Reduce the distribution clonign in
`ResolverOutput::from_state` by using an `Arc`.
The same profile excerpt for the resolver with the branch (note that
there is now an unrelated block between the two we optimized):
Wall times are noisy, but the profiles show those changes as
improvements.
```
$ hyperfine --warmup 2 "./uv-main pip compile --no-progress scripts/requirements/airflow.in --universal" "./uv-branch pip compile --no-progress scripts/requirements/airflow.in --universal"
Benchmark 1: ./uv-main pip compile --no-progress scripts/requirements/airflow.in --universal
Time (mean ± σ): 99.1 ms ± 3.8 ms [User: 111.8 ms, System: 115.5 ms]
Range (min … max): 93.6 ms … 110.4 ms 29 runs
Benchmark 2: ./uv-branch pip compile --no-progress scripts/requirements/airflow.in --universal
Time (mean ± σ): 97.1 ms ± 4.3 ms [User: 114.8 ms, System: 112.0 ms]
Range (min … max): 90.9 ms … 112.4 ms 29 runs
Summary
./uv-branch pip compile --no-progress scripts/requirements/airflow.in --universal ran
1.02 ± 0.06 times faster than ./uv-main pip compile --no-progress scripts/requirements/airflow.in --universal
```
Instead of always using all available threads for bytecode compilation,
respect `UV_CONCURRENT_INSTALLS`, so the parallelism is configurable
instead of hardcoded. We reuse the install limit since bytecode
compilation only runs after install.
Initially, we were limiting Git schemes to HTTPS and SSH as only
supported schemes. We lost this validation in #3429. This incidentally
allowed file schemes, which apparently work with Git out of the box.
A caveat for this is that in tool.uv.sources, we parse the git field
always as URL. This caused a problem with #11425: repo = { git =
'c:\path\to\repo', rev = "xxxxx" } was parsed as a URL where c: is the
scheme, causing a bad error message down the line.
This PR:
* Puts Git URL validation back in place. It bans everything but HTTPS,
SSH, and file URLs. This could be a breaking change, if users were using
a git transport protocol were not aware of, even though never
intentionally supported.
* Allows file: URL in Git: This seems to be supported by Git and we were
supporting it albeit unintentionally, so it's reasonable to continue to
support it.
* It does not allow relative paths in the git field in tool.uv.sources.
Absolute file URLs are supported, whether we want relative file URLs for
Git too should be discussed separately.
Closes#3429: We reject the input with a proper error message, while
hinting the user towards file:. If there's still desire for relative
path support, we can keep it open.
---------
Co-authored-by: Charlie Marsh <charlie.r.marsh@gmail.com>
We want to build `uv-build` without depending on the network crates. In
preparation for that, we split uv-git into uv-git and uv-git-types,
where only uv-git depends on reqwest, so that uv-build can use
uv-git-types.
## Summary
We currently enforce that if you do `uv pip install
./dist/iniconfig-1.0.0.tar.gz`, the build _must_ produce a wheel like
`iniconfig-1.0.0-py3-none-any.whl` (i.e., the name and version must
match). It turns out some packages produce a wheel that has a local
suffix on it, like `vllm`. This PR makes the check a little more
permissive in that we now accept `1.0.0` or that version with a local
suffix (e.g., `1.0.0+cpu`). I don't love this practice, but we already
relaxed this check when _installing_ a wheel, so this seems reasonable:
5e15881dcc/crates/uv-install-wheel/src/install.rs (L50-L52)
Note that this is _still_ stricter than pip. pip seems to only require
that the package name is the same (i.e., `iniconfig` matches
`iniconfig`; but they'll happily install a wheel like
`iniconfig-2.0.0-py3-none-any.whl` given
`./dist/iniconfig-1.0.0.tar.gz`).
Closes https://github.com/astral-sh/uv/issues/11038.
uv-install-wheel had the logic for laying out the installation and for
linking a directory in the same module. We split them up to isolate each
module's logic and tighten the crate's interface to only expose top
level members.
No logic changes, only moving code around.
Log the file that failed to bytecode compile when encountering a timeout
for debugging #6105 better.
[sysinfo](https://lib.rs/crates/sysinfo) would give us the option to
report memory usage too, but i'm hesitant to add a dependency just for
the error path.
## Summary
When `--upgrade` is provided, we should retain already-installed
packages _if_ they're newer than whatever is available from the
registry.
Closes https://github.com/astral-sh/uv/issues/10089.
## Summary
Sort of undecided on this. These are already stored as `dyn Reporter` in
each struct, so we're already using dynamic dispatch in that sense. But
all the methods take `impl Reporter`. This is sometimes nice (the
callsites are simpler?), but it also means that in practice, you often
_can't_ pass `None` to these methods that accept `Option<impl
Reporter>`, because Rust can't infer the generic type.
Anyway, this adds more consistency and simplifies the setup by using
`Arc<dyn Reporter>` everywhere.
Build failures are one of the most common user facing failures that
aren't "obivous" errors (such as typos) or resolver errors. Currently,
they show more technical details than being focussed on this being an
error in a subprocess that is either on the side of the package or -
more likely - in the build environment, e.g. the user needs to install a
dev package or their python version is incompatible.
The new error message clearly delineates the part that's important (this
is a build backend problem) from the internals (we called this hook) and
is consistent about which part of the dist building stage failed. We
have to calibrate the exact wording of the error message some more. Most
of the implementation is working around the orphan rule, (this)error
rules and trait rules, so it came out more of a refactoring than
intended.
Example:
The resolver methods are already too large and complex, especially
`choose_version*`, so i wanted to shrink and simplify them a bit before
adding new methods to them.
I've split `MetadataResponse` into three variants: success, non-fatal
error (reported through pubgrub), fatal error (reported as error trace).
The resulting non-fatal `MetadataUnavailable` type is equivalent to the
`IncompletePackage` type, so they are now merged. (`UnavailableVersion`
is a bit different since, besides the extra `IncompatibleDist` variant,
it have no error source attached). This shows that the missing metadata
variant was unused, which I removed.
Tagging as error messages for the logging format changes.
Instead of modifying the error to replace a dummy derivation chain from
construction with the real one, build the error with the real derivation
chain directly.
This came up when trying to improve the build error reporting.
Introduces `DistErrorKind` to avoid error variants for each case that
are only different in one line of the message.
## Summary
This change introduces the `UV_NO_INSTALLER_METADATA` environment
variable
as a way to opt out of the extra installer metadata files that `uv` is
creating.
This is important to achieve reproducible builds in distribution
packaging, allowing to replace usage of
[installer](https://pypi.org/project/installer) with `uv pip install`.
At the time of writing these files are:
- `uv_cache.json`
Contains timestamps which are non-reproducible.
These hashes also leak in to the `RECORD` file.
- `direct_url.json`
Contains the path to the installed wheel.
While not non-reproducible it's not required for distribution packaging.
- `INSTALLER`
Again, not non-reproducible, but of no value in distribution packaging.
## Test Plan
Automated test added.
---------
Co-authored-by: Charlie Marsh <charlie.r.marsh@gmail.com>
When performing a noop sync, we don't need the rayon threadpool, yet we
pay for its initialization:
Be making the initialization lazy, we avoid that cost:
This code runs every time before user code in `uv run`.
This means that before calling rayon, one now needs to call
`LazyLock::force(&RAYON_INITIALIZE);`.
Performance mode (CPU 0 is a perf core):
```
$ taskset -c 0 hyperfine --warmup 5 -N "/home/konsti/projects/uv/uv-main sync" "/home/konsti/projects/uv/target/profiling/uv sync"
Benchmark 1: /home/konsti/projects/uv/uv-main sync
Time (mean ± σ): 4.5 ms ± 0.1 ms [User: 2.7 ms, System: 1.8 ms]
Range (min … max): 4.4 ms … 6.4 ms 640 runs
Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet system without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.
Benchmark 2: /home/konsti/projects/uv/target/profiling/uv sync
Time (mean ± σ): 4.4 ms ± 0.1 ms [User: 2.7 ms, System: 1.6 ms]
Range (min … max): 4.3 ms … 5.0 ms 679 runs
Summary
/home/konsti/projects/uv/target/profiling/uv sync ran
1.03 ± 0.04 times faster than /home/konsti/projects/uv/uv-main sync
```
Power saver mode:
```
$ hyperfine --warmup 5 -N "/home/konsti/projects/uv/uv-main sync" "/home/konsti/projects/uv/target/profiling/uv sync"
Benchmark 1: /home/konsti/projects/uv/uv-main sync
Time (mean ± σ): 28.1 ms ± 1.2 ms [User: 15.5 ms, System: 20.3 ms]
Range (min … max): 25.7 ms … 31.9 ms 102 runs
Benchmark 2: /home/konsti/projects/uv/target/profiling/uv sync
Time (mean ± σ): 24.0 ms ± 1.2 ms [User: 13.8 ms, System: 9.9 ms]
Range (min … max): 22.2 ms … 28.2 ms 122 runs
Summary
/home/konsti/projects/uv/target/profiling/uv sync ran
1.17 ± 0.08 times faster than /home/konsti/projects/uv/uv-main sync
```
## 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
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.
## Summary
Shows similar diagnostics for failures that happen at install time,
rather than resolve time. This will ultimately feed into
https://github.com/astral-sh/uv/issues/8962 since we'll now have
consolidated handling for these kinds of failures.
## 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?
## 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"
```
## Summary
This PR declares and documents all environment variables that are used
in one way or another in `uv`, either internally, or externally, or
transitively under a common struct.
I think over time as uv has grown there's been many environment
variables introduced. Its harder to know which ones exists, which ones
are missing, what they're used for, or where are they used across the
code. The docs only documents a handful of them, for others you'd have
to dive into the code and inspect across crates to know which crates
they're used on or where they're relevant.
This PR is a starting attempt to unify them, make it easier to discover
which ones we have, and maybe unlock future posibilities in automating
generating documentation for them.
I think we can split out into multiple structs later to better organize,
but given the high influx of PR's and possibly new environment variables
introduced/re-used, it would be hard to try to organize them all now
into their proper namespaced struct while this is all happening given
merge conflicts and/or keeping up to date.
I don't think this has any impact on performance as they all should
still be inlined, although it may affect local build times on changes to
the environment vars as more crates would likely need a rebuild. Lastly,
some of them are declared but not used in the code, for example those in
`build.rs`. I left them declared because I still think it's useful to at
least have a reference.
Did I miss any? Are their initial docs cohesive?
Note, `uv-static` is a terrible name for a new crate, thoughts? Others
considered `uv-vars`, `uv-consts`.
## Test Plan
Existing tests
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).
## Summary
This is a longstanding piece of technical debt. After we resolve, we
have a bunch of `ResolvedDist` entries. We then convert those to
`Requirement` (which is lossy -- we lose information like "the index
that the package was resolved to"), and then back to `Dist`.
## Summary
Historically, we've allowed the use of wheels that were downloaded from
PyPI even when the user passes `--no-binary`, if the wheel exists in the
cache. This PR modifies the cache lookup code such that we respect
`--no-build` and `--no-binary` in those paths.
Closes https://github.com/astral-sh/uv/issues/2154.
## Summary
If `--config-settings` are provided, we cache the built wheels under one
more subdirectory.
We _don't_ invalidate the actual source (i.e., trigger a re-download) or
metadata, though -- those can be reused even when `--config-settings`
change.
Closes https://github.com/astral-sh/uv/issues/7028.
## Summary
This PR adds a more flexible cache invalidation abstraction for uv, and
uses that new abstraction to improve support for dynamic metadata.
Specifically, instead of relying solely on a timestamp, we now pass
around a `CacheInfo` struct which (as of now) contains
`Option<Timestamp>` and `Option<Commit>`. The `CacheInfo` is saved in
`dist-info` as `uv_cache.json`, so we can test already-installed
distributions for cache validity (along with testing _cached_
distributions for cache validity).
Beyond the defaults (`pyproject.toml`, `setup.py`, and `setup.cfg`
changes), users can also specify additional cache keys, and it's easy
for us to extend support in the future. Right now, cache keys can either
be instructions to include the current commit (for `setuptools_scm` and
similar) or file paths (for `hatch-requirements-txt` and similar):
```toml
[tool.uv]
cache-keys = [{ file = "requirements.txt" }, { git = true }]
```
This change should be fully backwards compatible.
Closes https://github.com/astral-sh/uv/issues/6964.
Closes https://github.com/astral-sh/uv/issues/6255.
Closes https://github.com/astral-sh/uv/issues/6860.
<!--
Thank you for contributing to uv! To help us out with reviewing, please
consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title?
- Does this pull request include references to any relevant issues?
-->
## Summary
Separate exceptions for different timeouts to make it easier to debug
issues like #6105.
<!-- What's the purpose of the change? What does it do, and why? -->
## Test Plan
<!-- How was it tested? -->
Not tested at all.
## Summary
Use a dedicated source type for non-package requirements. Also enables
us to support non-package `path` dependencies _and_ removes the need to
have the member `pyproject.toml` files available when we sync _and_
makes it explicit which dependencies are virtual vs. not (as evidenced
by the snapshot changes). All good things!
## 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.
For users who were using absolute paths in the `pyproject.toml`
previously, this is a behavior change: We now convert all absolute paths
in `path` entries to relative paths. Since i assume that no-one relies
on absolute path in their lockfiles - they are intended to be portable -
I'm tagging this as a bugfix.
Closes https://github.com/astral-sh/uv/pull/6438
Fixes https://github.com/astral-sh/uv/issues/6371
Right now, the URL gets out-of-sync with the install path, since the
install path is canonicalized. This leads to a subtle error on Windows
(in CI) in which we don't preserve caching across resolution and
installation.
