//! Build wheels from source distributions
//!
//!
use std::ffi::OsString;
use std::fmt::{Display, Formatter};
use std::io;
use std::path::{Path, PathBuf};
use std::process::{ExitStatus, Output};
use std::rc::Rc;
use std::str::FromStr;
use std::{env, iter};
use fs_err as fs;
use indoc::formatdoc;
use itertools::Itertools;
use once_cell::sync::Lazy;
use regex::Regex;
use rustc_hash::FxHashMap;
use serde::de::{value, SeqAccess, Visitor};
use serde::{de, Deserialize, Deserializer};
use tempfile::{tempdir_in, TempDir};
use thiserror::Error;
use tokio::process::Command;
use tokio::sync::{Mutex, Semaphore};
use tracing::{debug, info_span, instrument, Instrument};
use distribution_types::{Requirement, Resolution};
use pep440_rs::Version;
use pep508_rs::PackageName;
use pypi_types::VerbatimParsedUrl;
use uv_configuration::{BuildKind, ConfigSettings, SetupPyStrategy};
use uv_fs::{PythonExt, Simplified};
use uv_interpreter::{Interpreter, PythonEnvironment};
use uv_types::{BuildContext, BuildIsolation, SourceBuildTrait};
/// e.g. `pygraphviz/graphviz_wrap.c:3020:10: fatal error: graphviz/cgraph.h: No such file or directory`
static MISSING_HEADER_RE_GCC: Lazy = Lazy::new(|| {
Regex::new(
r".*\.(?:c|c..|h|h..):\d+:\d+: fatal error: (.*\.(?:h|h..)): No such file or directory",
)
.unwrap()
});
/// e.g. `pygraphviz/graphviz_wrap.c:3023:10: fatal error: 'graphviz/cgraph.h' file not found`
static MISSING_HEADER_RE_CLANG: Lazy = Lazy::new(|| {
Regex::new(r".*\.(?:c|c..|h|h..):\d+:\d+: fatal error: '(.*\.(?:h|h..))' file not found")
.unwrap()
});
/// e.g. `pygraphviz/graphviz_wrap.c(3023): fatal error C1083: Cannot open include file: 'graphviz/cgraph.h': No such file or directory`
static MISSING_HEADER_RE_MSVC: Lazy = Lazy::new(|| {
Regex::new(r".*\.(?:c|c..|h|h..)\(\d+\): fatal error C1083: Cannot open include file: '(.*\.(?:h|h..))': No such file or directory")
.unwrap()
});
/// e.g. `/usr/bin/ld: cannot find -lncurses: No such file or directory`
static LD_NOT_FOUND_RE: Lazy = Lazy::new(|| {
Regex::new(r"/usr/bin/ld: cannot find -l([a-zA-Z10-9]+): No such file or directory").unwrap()
});
/// e.g. `error: invalid command 'bdist_wheel'`
static WHEEL_NOT_FOUND_RE: Lazy =
Lazy::new(|| Regex::new(r"error: invalid command 'bdist_wheel'").unwrap());
/// The default backend to use when PEP 517 is used without a `build-system` section.
static DEFAULT_BACKEND: Lazy = Lazy::new(|| Pep517Backend {
backend: "setuptools.build_meta:__legacy__".to_string(),
backend_path: None,
requirements: vec![Requirement::from(
pep508_rs::Requirement::from_str("setuptools >= 40.8.0").unwrap(),
)],
});
/// The requirements for `--legacy-setup-py` builds.
static SETUP_PY_REQUIREMENTS: Lazy<[Requirement; 2]> = Lazy::new(|| {
[
Requirement::from(pep508_rs::Requirement::from_str("setuptools >= 40.8.0").unwrap()),
Requirement::from(pep508_rs::Requirement::from_str("wheel").unwrap()),
]
});
#[derive(Error, Debug)]
pub enum Error {
#[error(transparent)]
IO(#[from] io::Error),
#[error("Invalid source distribution: {0}")]
InvalidSourceDist(String),
#[error("Invalid `pyproject.toml`")]
InvalidPyprojectToml(#[from] toml::de::Error),
#[error("Editable installs with setup.py legacy builds are unsupported, please specify a build backend in pyproject.toml")]
EditableSetupPy,
#[error("Failed to install requirements from {0}")]
RequirementsInstall(&'static str, #[source] anyhow::Error),
#[error("Failed to create temporary virtualenv")]
Virtualenv(#[from] uv_virtualenv::Error),
#[error("Failed to run {0}")]
CommandFailed(PathBuf, #[source] io::Error),
#[error("{message} with {exit_code}\n--- stdout:\n{stdout}\n--- stderr:\n{stderr}\n---")]
BuildBackend {
message: String,
exit_code: ExitStatus,
stdout: String,
stderr: String,
},
/// Nudge the user towards installing the missing dev library
#[error("{message} with {exit_code}\n--- stdout:\n{stdout}\n--- stderr:\n{stderr}\n---")]
MissingHeader {
message: String,
exit_code: ExitStatus,
stdout: String,
stderr: String,
#[source]
missing_header_cause: MissingHeaderCause,
},
#[error("Failed to build PATH for build script")]
BuildScriptPath(#[source] env::JoinPathsError),
}
#[derive(Debug)]
enum MissingLibrary {
Header(String),
Linker(String),
PythonPackage(String),
}
#[derive(Debug, Error)]
pub struct MissingHeaderCause {
missing_library: MissingLibrary,
version_id: String,
}
impl Display for MissingHeaderCause {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
match &self.missing_library {
MissingLibrary::Header(header) => {
write!(
f,
"This error likely indicates that you need to install a library that provides \"{}\" for {}",
header, self.version_id
)
}
MissingLibrary::Linker(library) => {
write!(
f,
"This error likely indicates that you need to install the library that provides a shared library \
for {library} for {version_id} (e.g. lib{library}-dev)",
library = library, version_id = self.version_id
)
}
MissingLibrary::PythonPackage(package) => {
write!(
f,
"This error likely indicates that you need to `uv pip install {package}` into the build environment for {version_id}",
package = package, version_id = self.version_id
)
}
}
}
}
impl Error {
fn from_command_output(
message: String,
output: &Output,
version_id: impl Into,
) -> Self {
let stdout = String::from_utf8_lossy(&output.stdout).trim().to_string();
let stderr = String::from_utf8_lossy(&output.stderr).trim().to_string();
// In the cases i've seen it was the 5th and 3rd last line (see test case), 10 seems like a reasonable cutoff
let missing_library = stderr.lines().rev().take(10).find_map(|line| {
if let Some((_, [header])) = MISSING_HEADER_RE_GCC
.captures(line.trim())
.or(MISSING_HEADER_RE_CLANG.captures(line.trim()))
.or(MISSING_HEADER_RE_MSVC.captures(line.trim()))
.map(|c| c.extract())
{
Some(MissingLibrary::Header(header.to_string()))
} else if let Some((_, [library])) =
LD_NOT_FOUND_RE.captures(line.trim()).map(|c| c.extract())
{
Some(MissingLibrary::Linker(library.to_string()))
} else if WHEEL_NOT_FOUND_RE.is_match(line.trim()) {
Some(MissingLibrary::PythonPackage("wheel".to_string()))
} else {
None
}
});
if let Some(missing_library) = missing_library {
return Self::MissingHeader {
message,
exit_code: output.status,
stdout,
stderr,
missing_header_cause: MissingHeaderCause {
missing_library,
version_id: version_id.into(),
},
};
}
Self::BuildBackend {
message,
exit_code: output.status,
stdout,
stderr,
}
}
}
/// A `pyproject.toml` as specified in PEP 517.
#[derive(Deserialize, Debug, Clone, PartialEq, Eq)]
#[serde(rename_all = "kebab-case")]
struct PyProjectToml {
/// Build-related data
build_system: Option,
/// Project metadata
project: Option,
}
/// The `[project]` section of a pyproject.toml as specified in PEP 621.
///
/// This representation only includes a subset of the fields defined in PEP 621 necessary for
/// informing wheel builds.
#[derive(Deserialize, Debug, Clone, PartialEq, Eq)]
#[serde(rename_all = "kebab-case")]
struct Project {
/// The name of the project
name: PackageName,
/// The version of the project as supported by PEP 440
version: Option,
/// Specifies which fields listed by PEP 621 were intentionally unspecified so another tool
/// can/will provide such metadata dynamically.
dynamic: Option>,
}
/// The `[build-system]` section of a pyproject.toml as specified in PEP 517.
#[derive(Deserialize, Debug, Clone, PartialEq, Eq)]
#[serde(rename_all = "kebab-case")]
struct BuildSystem {
/// PEP 508 dependencies required to execute the build system.
requires: Vec>,
/// A string naming a Python object that will be used to perform the build.
build_backend: Option,
/// Specify that their backend code is hosted in-tree, this key contains a list of directories.
backend_path: Option,
}
impl BackendPath {
/// Return an iterator over the paths in the backend path.
fn iter(&self) -> impl Iterator {
self.0.iter().map(String::as_str)
}
}
#[derive(Debug, Clone, PartialEq, Eq)]
struct BackendPath(Vec);
impl<'de> Deserialize<'de> for BackendPath {
fn deserialize(deserializer: D) -> Result
where
D: Deserializer<'de>,
{
struct StringOrVec;
impl<'de> Visitor<'de> for StringOrVec {
type Value = Vec;
fn expecting(&self, formatter: &mut Formatter) -> std::fmt::Result {
formatter.write_str("list of strings")
}
fn visit_str(self, s: &str) -> Result
where
E: de::Error,
{
// Allow exactly `backend-path = "."`, as used in `flit_core==2.3.0`.
if s == "." {
Ok(vec![".".to_string()])
} else {
Err(de::Error::invalid_value(de::Unexpected::Str(s), &self))
}
}
fn visit_seq(self, seq: S) -> Result
where
S: SeqAccess<'de>,
{
Deserialize::deserialize(value::SeqAccessDeserializer::new(seq))
}
}
deserializer.deserialize_any(StringOrVec).map(BackendPath)
}
}
/// `[build-backend]` from pyproject.toml
#[derive(Debug, Clone, PartialEq, Eq)]
struct Pep517Backend {
/// The build backend string such as `setuptools.build_meta:__legacy__` or `maturin` from
/// `build-backend.backend` in pyproject.toml
///
///
backend: String,
/// `build-backend.requirements` in pyproject.toml
requirements: Vec,
///
backend_path: Option,
}
impl Pep517Backend {
fn backend_import(&self) -> String {
let import = if let Some((path, object)) = self.backend.split_once(':') {
format!("from {path} import {object} as backend")
} else {
format!("import {} as backend", self.backend)
};
let backend_path_encoded = self
.backend_path
.iter()
.flat_map(BackendPath::iter)
.map(|path| {
// Turn into properly escaped python string
'"'.to_string()
+ &path.replace('\\', "\\\\").replace('"', "\\\"")
+ &'"'.to_string()
})
.join(", ");
// > Projects can specify that their backend code is hosted in-tree by including the
// > backend-path key in pyproject.toml. This key contains a list of directories, which the
// > frontend will add to the start of sys.path when loading the backend, and running the
// > backend hooks.
formatdoc! {r#"
import sys
if sys.path[0] == "":
sys.path.pop(0)
sys.path = [{backend_path}] + sys.path
{import}
"#, backend_path = backend_path_encoded}
}
}
/// Uses an [`Rc`] internally, clone freely.
#[derive(Debug, Default, Clone)]
pub struct SourceBuildContext {
/// An in-memory resolution of the default backend's requirements for PEP 517 builds.
default_resolution: Rc>>,
/// An in-memory resolution of the build requirements for `--legacy-setup-py` builds.
setup_py_resolution: Rc>>,
}
/// Holds the state through a series of PEP 517 frontend to backend calls or a single setup.py
/// invocation.
///
/// This keeps both the temp dir and the result of a potential `prepare_metadata_for_build_wheel`
/// call which changes how we call `build_wheel`.
pub struct SourceBuild {
temp_dir: TempDir,
source_tree: PathBuf,
config_settings: ConfigSettings,
/// If performing a PEP 517 build, the backend to use.
pep517_backend: Option,
/// The PEP 621 project metadata, if any.
project: Option,
/// The virtual environment in which to build the source distribution.
venv: PythonEnvironment,
/// Populated if `prepare_metadata_for_build_wheel` was called.
///
/// > If the build frontend has previously called `prepare_metadata_for_build_wheel` and depends
/// > on the wheel resulting from this call to have metadata matching this earlier call, then
/// > it should provide the path to the created .dist-info directory as the `metadata_directory`
/// > argument. If this argument is provided, then `build_wheel` MUST produce a wheel with
/// > identical metadata. The directory passed in by the build frontend MUST be identical to the
/// > directory created by `prepare_metadata_for_build_wheel`, including any unrecognized files
/// > it created.
metadata_directory: Option,
/// Package id such as `foo-1.2.3`, for error reporting
version_id: String,
/// Whether we do a regular PEP 517 build or an PEP 660 editable build
build_kind: BuildKind,
/// Modified PATH that contains the `venv_bin`, `user_path` and `system_path` variables in that order
modified_path: OsString,
/// Environment variables to be passed in during metadata or wheel building
environment_variables: FxHashMap,
/// Runner for Python scripts.
runner: PythonRunner,
}
impl SourceBuild {
/// Create a virtual environment in which to build a source distribution, extracting the
/// contents from an archive if necessary.
///
/// `source_dist` is for error reporting only.
#[allow(clippy::too_many_arguments)]
pub async fn setup(
source: &Path,
subdirectory: Option<&Path>,
interpreter: &Interpreter,
build_context: &impl BuildContext,
source_build_context: SourceBuildContext,
version_id: String,
setup_py: SetupPyStrategy,
config_settings: ConfigSettings,
build_isolation: BuildIsolation<'_>,
build_kind: BuildKind,
mut environment_variables: FxHashMap,
concurrent_builds: usize,
) -> Result {
let temp_dir = tempdir_in(build_context.cache().root())?;
let source_tree = if let Some(subdir) = subdirectory {
source.join(subdir)
} else {
source.to_path_buf()
};
let default_backend: Pep517Backend = DEFAULT_BACKEND.clone();
// Check if we have a PEP 517 build backend.
let (pep517_backend, project) =
Self::extract_pep517_backend(&source_tree, setup_py, &default_backend)
.map_err(|err| *err)?;
// Create a virtual environment, or install into the shared environment if requested.
let venv = match build_isolation {
BuildIsolation::Isolated => uv_virtualenv::create_venv(
&temp_dir.path().join(".venv"),
interpreter.clone(),
uv_virtualenv::Prompt::None,
false,
false,
)?,
BuildIsolation::Shared(venv) => venv.clone(),
};
// Setup the build environment. If build isolation is disabled, we assume the build
// environment is already setup.
if build_isolation.is_isolated() {
let resolved_requirements = Self::get_resolved_requirements(
build_context,
source_build_context,
&default_backend,
pep517_backend.as_ref(),
)
.await?;
build_context
.install(&resolved_requirements, &venv)
.await
.map_err(|err| {
Error::RequirementsInstall("build-system.requires (install)", err)
})?;
}
// Figure out what the modified path should be
// Remove the PATH variable from the environment variables if it's there
let user_path = environment_variables.remove(&OsString::from("PATH"));
// See if there is an OS PATH variable
let os_path = env::var_os("PATH");
// Prepend the user supplied PATH to the existing OS PATH
let modified_path = if let Some(user_path) = user_path {
match os_path {
// Prepend the user supplied PATH to the existing PATH
Some(env_path) => {
let user_path = PathBuf::from(user_path);
let new_path = env::split_paths(&user_path).chain(env::split_paths(&env_path));
Some(env::join_paths(new_path).map_err(Error::BuildScriptPath)?)
}
// Use the user supplied PATH
None => Some(user_path),
}
} else {
os_path
};
// Prepend the venv bin directory to the modified path
let modified_path = if let Some(path) = modified_path {
let venv_path = iter::once(venv.scripts().to_path_buf()).chain(env::split_paths(&path));
env::join_paths(venv_path).map_err(Error::BuildScriptPath)?
} else {
OsString::from(venv.scripts())
};
// Create the PEP 517 build environment. If build isolation is disabled, we assume the build
// environment is already setup.
let runner = PythonRunner::new(concurrent_builds);
if build_isolation.is_isolated() {
if let Some(pep517_backend) = &pep517_backend {
create_pep517_build_environment(
&runner,
&source_tree,
&venv,
pep517_backend,
build_context,
&version_id,
build_kind,
&config_settings,
&environment_variables,
&modified_path,
&temp_dir,
)
.await?;
}
}
Ok(Self {
temp_dir,
source_tree,
pep517_backend,
project,
venv,
build_kind,
config_settings,
metadata_directory: None,
version_id,
environment_variables,
modified_path,
runner,
})
}
async fn get_resolved_requirements(
build_context: &impl BuildContext,
source_build_context: SourceBuildContext,
default_backend: &Pep517Backend,
pep517_backend: Option<&Pep517Backend>,
) -> Result {
Ok(if let Some(pep517_backend) = pep517_backend {
if pep517_backend.requirements == default_backend.requirements {
let mut resolution = source_build_context.default_resolution.lock().await;
if let Some(resolved_requirements) = &*resolution {
resolved_requirements.clone()
} else {
let resolved_requirements = build_context
.resolve(&default_backend.requirements)
.await
.map_err(|err| {
Error::RequirementsInstall("setup.py build (resolve)", err)
})?;
*resolution = Some(resolved_requirements.clone());
resolved_requirements
}
} else {
build_context
.resolve(&pep517_backend.requirements)
.await
.map_err(|err| {
Error::RequirementsInstall("build-system.requires (resolve)", err)
})?
}
} else {
// Install default requirements for `setup.py`-based builds.
let mut resolution = source_build_context.setup_py_resolution.lock().await;
if let Some(resolved_requirements) = &*resolution {
resolved_requirements.clone()
} else {
let resolved_requirements = build_context
.resolve(&*SETUP_PY_REQUIREMENTS)
.await
.map_err(|err| Error::RequirementsInstall("setup.py build (resolve)", err))?;
*resolution = Some(resolved_requirements.clone());
resolved_requirements
}
})
}
/// Extract the PEP 517 backend from the `pyproject.toml` or `setup.py` file.
fn extract_pep517_backend(
source_tree: &Path,
setup_py: SetupPyStrategy,
default_backend: &Pep517Backend,
) -> Result<(Option, Option), Box> {
match fs::read_to_string(source_tree.join("pyproject.toml")) {
Ok(toml) => {
let pyproject_toml: PyProjectToml =
toml::from_str(&toml).map_err(Error::InvalidPyprojectToml)?;
let backend = if let Some(build_system) = pyproject_toml.build_system {
Pep517Backend {
// If `build-backend` is missing, inject the legacy setuptools backend, but
// retain the `requires`, to match `pip` and `build`. Note that while PEP 517
// says that in this case we "should revert to the legacy behaviour of running
// `setup.py` (either directly, or by implicitly invoking the
// `setuptools.build_meta:__legacy__` backend)", we found that in practice, only
// the legacy setuptools backend is allowed. See also:
// https://github.com/pypa/build/blob/de5b44b0c28c598524832dff685a98d5a5148c44/src/build/__init__.py#L114-L118
backend: build_system
.build_backend
.unwrap_or_else(|| "setuptools.build_meta:__legacy__".to_string()),
backend_path: build_system.backend_path,
requirements: build_system
.requires
.into_iter()
.map(Requirement::from)
.collect(),
}
} else {
// If a `pyproject.toml` is present, but `[build-system]` is missing, proceed with
// a PEP 517 build using the default backend, to match `pip` and `build`.
default_backend.clone()
};
Ok((Some(backend), pyproject_toml.project))
}
Err(err) if err.kind() == io::ErrorKind::NotFound => {
// We require either a `pyproject.toml` or a `setup.py` file at the top level.
if !source_tree.join("setup.py").is_file() {
return Err(Box::new(Error::InvalidSourceDist(
"The archive contains neither a `pyproject.toml` nor a `setup.py` file at the top level"
.to_string(),
)));
}
// If no `pyproject.toml` is present, by default, proceed with a PEP 517 build using
// the default backend, to match `build`. `pip` uses `setup.py` directly in this
// case (which we allow via `SetupPyStrategy::Setuptools`), but plans to make PEP
// 517 builds the default in the future.
// See: https://github.com/pypa/pip/issues/9175.
match setup_py {
SetupPyStrategy::Pep517 => Ok((Some(default_backend.clone()), None)),
SetupPyStrategy::Setuptools => Ok((None, None)),
}
}
Err(err) => Err(Box::new(err.into())),
}
}
/// Try calling `prepare_metadata_for_build_wheel` to get the metadata without executing the
/// actual build.
pub async fn get_metadata_without_build(&mut self) -> Result