//! Build wheels from source distributions
//!
//!
use std::ffi::OsString;
use std::fmt::{Display, Formatter};
use std::io;
use std::io::BufRead;
use std::path::{Path, PathBuf};
use std::process::Output;
use std::str::FromStr;
use std::sync::Arc;
use std::{env, iter};
use fs_err as fs;
use indoc::formatdoc;
use itertools::Itertools;
use once_cell::sync::Lazy;
use pyproject_toml::Project;
use regex::Regex;
use rustc_hash::FxHashMap;
use serde::de::{value, SeqAccess, Visitor};
use serde::{de, Deserialize, Deserializer, Serialize};
use tempfile::{tempdir_in, TempDir};
use thiserror::Error;
use tokio::process::Command;
use tokio::sync::Mutex;
use tracing::{debug, info_span, instrument, Instrument};
use distribution_types::Resolution;
use pep508_rs::Requirement;
use uv_fs::Simplified;
use uv_interpreter::{Interpreter, PythonEnvironment};
use uv_traits::{BuildContext, BuildKind, ConfigSettings, SetupPyStrategy, SourceBuildTrait};
/// e.g. `pygraphviz/graphviz_wrap.c:3020:10: fatal error: graphviz/cgraph.h: No such file or directory`
static MISSING_HEADER_RE: Lazy = Lazy::new(|| {
Regex::new(
r".*\.(?:c|c..|h|h..):\d+:\d+: fatal error: (.*\.(?: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()
});
/// 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_str("wheel").unwrap(),
Requirement::from_str("setuptools >= 40.8.0").unwrap(),
],
});
#[derive(Error, Debug)]
pub enum Error {
#[error(transparent)]
IO(#[from] io::Error),
#[error("Failed to extract archive: {0}")]
Extraction(PathBuf, #[source] uv_extract::Error),
#[error("Unsupported archive format (extension not recognized): {0}")]
UnsupportedArchiveType(String),
#[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("Source distribution not found at: {0}")]
NotFound(PathBuf),
#[error("Failed to create temporary virtualenv")]
Virtualenv(#[from] uv_virtualenv::Error),
#[error("Failed to run {0}")]
CommandFailed(PathBuf, #[source] io::Error),
#[error("{message}:\n--- stdout:\n{stdout}\n--- stderr:\n{stderr}\n---")]
BuildBackend {
message: String,
stdout: String,
stderr: String,
},
/// Nudge the user towards installing the missing dev library
#[error("{message}:\n--- stdout:\n{stdout}\n--- stderr:\n{stderr}\n---")]
MissingHeader {
message: String,
stdout: String,
stderr: String,
#[source]
missing_header_cause: MissingHeaderCause,
},
#[error("Failed to build PATH for build script")]
BuildScriptPath(#[source] env::JoinPathsError),
}
#[derive(Debug)]
pub enum MissingLibrary {
Header(String),
Linker(String),
}
#[derive(Debug, Error)]
pub struct MissingHeaderCause {
missing_library: MissingLibrary,
package_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.package_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 {package_id} (e.g. lib{library}-dev)",
library = library, package_id = self.package_id
)
}
}
}
}
impl Error {
fn from_command_output(
message: String,
output: &Output,
package_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.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 {
None
}
});
if let Some(missing_library) = missing_library {
return Self::MissingHeader {
message,
stdout,
stderr,
missing_header_cause: MissingHeaderCause {
missing_library,
package_id: package_id.into(),
},
};
}
Self::BuildBackend {
message,
stdout,
stderr,
}
}
}
/// A pyproject.toml as specified in PEP 517
#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
#[serde(rename_all = "kebab-case")]
pub struct PyProjectToml {
/// Build-related data
pub build_system: Option,
/// Project metadata
pub project: Option,
}
/// The `[build-system]` section of a pyproject.toml as specified in PEP 517.
#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
#[serde(rename_all = "kebab-case")]
pub struct BuildSystem {
/// PEP 508 dependencies required to execute the build system.
pub requires: Vec,
/// A string naming a Python object that will be used to perform the build.
pub build_backend: Option,
/// Specify that their backend code is hosted in-tree, this key contains a list of directories.
pub 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(Serialize, Debug, Clone, PartialEq, Eq)]
#[serde(rename_all = "kebab-case")]
pub 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 [`Arc`] internally, clone freely
#[derive(Debug, Default, Clone)]
pub struct SourceBuildContext {
/// Cache the first resolution of `pip`, `setuptools` and `wheel` we made for setup.py (and
/// some PEP 517) builds so we can reuse it.
setup_py_resolution: Arc>>,
}
/// 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 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
package_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,
}
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,
package_id: String,
setup_py: SetupPyStrategy,
config_settings: ConfigSettings,
build_kind: BuildKind,
mut environment_variables: FxHashMap,
) -> Result {
let temp_dir = tempdir_in(build_context.cache().root())?;
let metadata = match fs::metadata(source) {
Ok(metadata) => metadata,
Err(err) if err.kind() == io::ErrorKind::NotFound => {
return Err(Error::NotFound(source.to_path_buf()));
}
Err(err) => return Err(err.into()),
};
let source_root = if metadata.is_dir() {
source.to_path_buf()
} else {
debug!("Unpacking for build: {}", source.display());
let extracted = temp_dir.path().join("extracted");
// Unzip the archive into the temporary directory.
let reader = fs_err::tokio::File::open(source).await?;
uv_extract::stream::archive(tokio::io::BufReader::new(reader), source, &extracted)
.await
.map_err(|err| Error::Extraction(extracted.clone(), err))?;
// Extract the top-level directory from the archive.
match uv_extract::strip_component(&extracted) {
Ok(top_level) => top_level,
Err(uv_extract::Error::NonSingularArchive(_)) => extracted,
Err(err) => return Err(Error::Extraction(extracted.clone(), err)),
}
};
let source_tree = if let Some(subdir) = subdirectory {
source_root.join(subdir)
} else {
source_root
};
let default_backend: Pep517Backend = DEFAULT_BACKEND.clone();
// Check if we have a PEP 517 build backend.
let pep517_backend = Self::get_pep517_backend(setup_py, &source_tree, &default_backend)
.map_err(|err| *err)?;
let venv = uv_virtualenv::create_venv(
&temp_dir.path().join(".venv"),
interpreter.clone(),
uv_virtualenv::Prompt::None,
false,
Vec::new(),
)?;
// Setup the build environment.
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())
};
if let Some(pep517_backend) = &pep517_backend {
create_pep517_build_environment(
&source_tree,
&venv,
pep517_backend,
build_context,
&package_id,
build_kind,
&config_settings,
&environment_variables,
&modified_path,
)
.await?;
}
Ok(Self {
temp_dir,
source_tree,
pep517_backend,
venv,
build_kind,
config_settings,
metadata_directory: None,
package_id,
environment_variables,
modified_path,
})
}
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.setup_py_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(&default_backend.requirements)
.await
.map_err(|err| Error::RequirementsInstall("setup.py build (resolve)", err))?;
*resolution = Some(resolved_requirements.clone());
resolved_requirements
}
})
}
fn get_pep517_backend(
setup_py: SetupPyStrategy,
source_tree: &Path,
default_backend: &Pep517Backend,
) -> Result