mirror of
https://git.proxmox.com/git/rustc
synced 2025-10-21 22:58:04 +00:00
1273 lines
40 KiB
Rust
1273 lines
40 KiB
Rust
//! A wrapper around the procedural macro API of the compiler's [`proc_macro`]
|
|
//! crate. This library serves two purposes:
|
|
//!
|
|
//! [`proc_macro`]: https://doc.rust-lang.org/proc_macro/
|
|
//!
|
|
//! - **Bring proc-macro-like functionality to other contexts like build.rs and
|
|
//! main.rs.** Types from `proc_macro` are entirely specific to procedural
|
|
//! macros and cannot ever exist in code outside of a procedural macro.
|
|
//! Meanwhile `proc_macro2` types may exist anywhere including non-macro code.
|
|
//! By developing foundational libraries like [syn] and [quote] against
|
|
//! `proc_macro2` rather than `proc_macro`, the procedural macro ecosystem
|
|
//! becomes easily applicable to many other use cases and we avoid
|
|
//! reimplementing non-macro equivalents of those libraries.
|
|
//!
|
|
//! - **Make procedural macros unit testable.** As a consequence of being
|
|
//! specific to procedural macros, nothing that uses `proc_macro` can be
|
|
//! executed from a unit test. In order for helper libraries or components of
|
|
//! a macro to be testable in isolation, they must be implemented using
|
|
//! `proc_macro2`.
|
|
//!
|
|
//! [syn]: https://github.com/dtolnay/syn
|
|
//! [quote]: https://github.com/dtolnay/quote
|
|
//!
|
|
//! # Usage
|
|
//!
|
|
//! The skeleton of a typical procedural macro typically looks like this:
|
|
//!
|
|
//! ```
|
|
//! extern crate proc_macro;
|
|
//!
|
|
//! # const IGNORE: &str = stringify! {
|
|
//! #[proc_macro_derive(MyDerive)]
|
|
//! # };
|
|
//! # #[cfg(wrap_proc_macro)]
|
|
//! pub fn my_derive(input: proc_macro::TokenStream) -> proc_macro::TokenStream {
|
|
//! let input = proc_macro2::TokenStream::from(input);
|
|
//!
|
|
//! let output: proc_macro2::TokenStream = {
|
|
//! /* transform input */
|
|
//! # input
|
|
//! };
|
|
//!
|
|
//! proc_macro::TokenStream::from(output)
|
|
//! }
|
|
//! ```
|
|
//!
|
|
//! If parsing with [Syn], you'll use [`parse_macro_input!`] instead to
|
|
//! propagate parse errors correctly back to the compiler when parsing fails.
|
|
//!
|
|
//! [`parse_macro_input!`]: https://docs.rs/syn/1.0/syn/macro.parse_macro_input.html
|
|
//!
|
|
//! # Unstable features
|
|
//!
|
|
//! The default feature set of proc-macro2 tracks the most recent stable
|
|
//! compiler API. Functionality in `proc_macro` that is not yet stable is not
|
|
//! exposed by proc-macro2 by default.
|
|
//!
|
|
//! To opt into the additional APIs available in the most recent nightly
|
|
//! compiler, the `procmacro2_semver_exempt` config flag must be passed to
|
|
//! rustc. We will polyfill those nightly-only APIs back to Rust 1.31.0. As
|
|
//! these are unstable APIs that track the nightly compiler, minor versions of
|
|
//! proc-macro2 may make breaking changes to them at any time.
|
|
//!
|
|
//! ```sh
|
|
//! RUSTFLAGS='--cfg procmacro2_semver_exempt' cargo build
|
|
//! ```
|
|
//!
|
|
//! Note that this must not only be done for your crate, but for any crate that
|
|
//! depends on your crate. This infectious nature is intentional, as it serves
|
|
//! as a reminder that you are outside of the normal semver guarantees.
|
|
//!
|
|
//! Semver exempt methods are marked as such in the proc-macro2 documentation.
|
|
//!
|
|
//! # Thread-Safety
|
|
//!
|
|
//! Most types in this crate are `!Sync` because the underlying compiler
|
|
//! types make use of thread-local memory, meaning they cannot be accessed from
|
|
//! a different thread.
|
|
|
|
// Proc-macro2 types in rustdoc of other crates get linked to here.
|
|
#![doc(html_root_url = "https://docs.rs/proc-macro2/1.0.27")]
|
|
#![cfg_attr(any(proc_macro_span, super_unstable), feature(proc_macro_span))]
|
|
#![cfg_attr(super_unstable, feature(proc_macro_raw_ident, proc_macro_def_site))]
|
|
#![cfg_attr(doc_cfg, feature(doc_cfg))]
|
|
#![allow(clippy::needless_doctest_main, clippy::vec_init_then_push)]
|
|
|
|
#[cfg(use_proc_macro)]
|
|
extern crate proc_macro;
|
|
|
|
mod marker;
|
|
mod parse;
|
|
|
|
#[cfg(wrap_proc_macro)]
|
|
mod detection;
|
|
|
|
// Public for proc_macro2::fallback::force() and unforce(), but those are quite
|
|
// a niche use case so we omit it from rustdoc.
|
|
#[doc(hidden)]
|
|
pub mod fallback;
|
|
|
|
#[cfg(not(wrap_proc_macro))]
|
|
use crate::fallback as imp;
|
|
#[path = "wrapper.rs"]
|
|
#[cfg(wrap_proc_macro)]
|
|
mod imp;
|
|
|
|
use crate::marker::Marker;
|
|
use std::cmp::Ordering;
|
|
use std::error::Error;
|
|
use std::fmt::{self, Debug, Display};
|
|
use std::hash::{Hash, Hasher};
|
|
use std::iter::FromIterator;
|
|
use std::ops::RangeBounds;
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
use std::path::PathBuf;
|
|
use std::str::FromStr;
|
|
|
|
/// An abstract stream of tokens, or more concretely a sequence of token trees.
|
|
///
|
|
/// This type provides interfaces for iterating over token trees and for
|
|
/// collecting token trees into one stream.
|
|
///
|
|
/// Token stream is both the input and output of `#[proc_macro]`,
|
|
/// `#[proc_macro_attribute]` and `#[proc_macro_derive]` definitions.
|
|
#[derive(Clone)]
|
|
pub struct TokenStream {
|
|
inner: imp::TokenStream,
|
|
_marker: Marker,
|
|
}
|
|
|
|
/// Error returned from `TokenStream::from_str`.
|
|
pub struct LexError {
|
|
inner: imp::LexError,
|
|
_marker: Marker,
|
|
}
|
|
|
|
impl TokenStream {
|
|
fn _new(inner: imp::TokenStream) -> TokenStream {
|
|
TokenStream {
|
|
inner,
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
fn _new_stable(inner: fallback::TokenStream) -> TokenStream {
|
|
TokenStream {
|
|
inner: inner.into(),
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
/// Returns an empty `TokenStream` containing no token trees.
|
|
pub fn new() -> TokenStream {
|
|
TokenStream::_new(imp::TokenStream::new())
|
|
}
|
|
|
|
/// Checks if this `TokenStream` is empty.
|
|
pub fn is_empty(&self) -> bool {
|
|
self.inner.is_empty()
|
|
}
|
|
}
|
|
|
|
/// `TokenStream::default()` returns an empty stream,
|
|
/// i.e. this is equivalent with `TokenStream::new()`.
|
|
impl Default for TokenStream {
|
|
fn default() -> Self {
|
|
TokenStream::new()
|
|
}
|
|
}
|
|
|
|
/// Attempts to break the string into tokens and parse those tokens into a token
|
|
/// stream.
|
|
///
|
|
/// May fail for a number of reasons, for example, if the string contains
|
|
/// unbalanced delimiters or characters not existing in the language.
|
|
///
|
|
/// NOTE: Some errors may cause panics instead of returning `LexError`. We
|
|
/// reserve the right to change these errors into `LexError`s later.
|
|
impl FromStr for TokenStream {
|
|
type Err = LexError;
|
|
|
|
fn from_str(src: &str) -> Result<TokenStream, LexError> {
|
|
let e = src.parse().map_err(|e| LexError {
|
|
inner: e,
|
|
_marker: Marker,
|
|
})?;
|
|
Ok(TokenStream::_new(e))
|
|
}
|
|
}
|
|
|
|
#[cfg(use_proc_macro)]
|
|
impl From<proc_macro::TokenStream> for TokenStream {
|
|
fn from(inner: proc_macro::TokenStream) -> TokenStream {
|
|
TokenStream::_new(inner.into())
|
|
}
|
|
}
|
|
|
|
#[cfg(use_proc_macro)]
|
|
impl From<TokenStream> for proc_macro::TokenStream {
|
|
fn from(inner: TokenStream) -> proc_macro::TokenStream {
|
|
inner.inner.into()
|
|
}
|
|
}
|
|
|
|
impl From<TokenTree> for TokenStream {
|
|
fn from(token: TokenTree) -> Self {
|
|
TokenStream::_new(imp::TokenStream::from(token))
|
|
}
|
|
}
|
|
|
|
impl Extend<TokenTree> for TokenStream {
|
|
fn extend<I: IntoIterator<Item = TokenTree>>(&mut self, streams: I) {
|
|
self.inner.extend(streams)
|
|
}
|
|
}
|
|
|
|
impl Extend<TokenStream> for TokenStream {
|
|
fn extend<I: IntoIterator<Item = TokenStream>>(&mut self, streams: I) {
|
|
self.inner
|
|
.extend(streams.into_iter().map(|stream| stream.inner))
|
|
}
|
|
}
|
|
|
|
/// Collects a number of token trees into a single stream.
|
|
impl FromIterator<TokenTree> for TokenStream {
|
|
fn from_iter<I: IntoIterator<Item = TokenTree>>(streams: I) -> Self {
|
|
TokenStream::_new(streams.into_iter().collect())
|
|
}
|
|
}
|
|
impl FromIterator<TokenStream> for TokenStream {
|
|
fn from_iter<I: IntoIterator<Item = TokenStream>>(streams: I) -> Self {
|
|
TokenStream::_new(streams.into_iter().map(|i| i.inner).collect())
|
|
}
|
|
}
|
|
|
|
/// Prints the token stream as a string that is supposed to be losslessly
|
|
/// convertible back into the same token stream (modulo spans), except for
|
|
/// possibly `TokenTree::Group`s with `Delimiter::None` delimiters and negative
|
|
/// numeric literals.
|
|
impl Display for TokenStream {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Display::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
/// Prints token in a form convenient for debugging.
|
|
impl Debug for TokenStream {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
impl LexError {
|
|
pub fn span(&self) -> Span {
|
|
Span::_new(self.inner.span())
|
|
}
|
|
}
|
|
|
|
impl Debug for LexError {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
impl Display for LexError {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Display::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
impl Error for LexError {}
|
|
|
|
/// The source file of a given `Span`.
|
|
///
|
|
/// This type is semver exempt and not exposed by default.
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(procmacro2_semver_exempt)))]
|
|
#[derive(Clone, PartialEq, Eq)]
|
|
pub struct SourceFile {
|
|
inner: imp::SourceFile,
|
|
_marker: Marker,
|
|
}
|
|
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
impl SourceFile {
|
|
fn _new(inner: imp::SourceFile) -> Self {
|
|
SourceFile {
|
|
inner,
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
/// Get the path to this source file.
|
|
///
|
|
/// ### Note
|
|
///
|
|
/// If the code span associated with this `SourceFile` was generated by an
|
|
/// external macro, this may not be an actual path on the filesystem. Use
|
|
/// [`is_real`] to check.
|
|
///
|
|
/// Also note that even if `is_real` returns `true`, if
|
|
/// `--remap-path-prefix` was passed on the command line, the path as given
|
|
/// may not actually be valid.
|
|
///
|
|
/// [`is_real`]: #method.is_real
|
|
pub fn path(&self) -> PathBuf {
|
|
self.inner.path()
|
|
}
|
|
|
|
/// Returns `true` if this source file is a real source file, and not
|
|
/// generated by an external macro's expansion.
|
|
pub fn is_real(&self) -> bool {
|
|
self.inner.is_real()
|
|
}
|
|
}
|
|
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
impl Debug for SourceFile {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
/// A line-column pair representing the start or end of a `Span`.
|
|
///
|
|
/// This type is semver exempt and not exposed by default.
|
|
#[cfg(span_locations)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(feature = "span-locations")))]
|
|
#[derive(Copy, Clone, Debug, PartialEq, Eq)]
|
|
pub struct LineColumn {
|
|
/// The 1-indexed line in the source file on which the span starts or ends
|
|
/// (inclusive).
|
|
pub line: usize,
|
|
/// The 0-indexed column (in UTF-8 characters) in the source file on which
|
|
/// the span starts or ends (inclusive).
|
|
pub column: usize,
|
|
}
|
|
|
|
#[cfg(span_locations)]
|
|
impl Ord for LineColumn {
|
|
fn cmp(&self, other: &Self) -> Ordering {
|
|
self.line
|
|
.cmp(&other.line)
|
|
.then(self.column.cmp(&other.column))
|
|
}
|
|
}
|
|
|
|
#[cfg(span_locations)]
|
|
impl PartialOrd for LineColumn {
|
|
fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
|
|
Some(self.cmp(other))
|
|
}
|
|
}
|
|
|
|
/// A region of source code, along with macro expansion information.
|
|
#[derive(Copy, Clone)]
|
|
pub struct Span {
|
|
inner: imp::Span,
|
|
_marker: Marker,
|
|
}
|
|
|
|
impl Span {
|
|
fn _new(inner: imp::Span) -> Span {
|
|
Span {
|
|
inner,
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
fn _new_stable(inner: fallback::Span) -> Span {
|
|
Span {
|
|
inner: inner.into(),
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
/// The span of the invocation of the current procedural macro.
|
|
///
|
|
/// Identifiers created with this span will be resolved as if they were
|
|
/// written directly at the macro call location (call-site hygiene) and
|
|
/// other code at the macro call site will be able to refer to them as well.
|
|
pub fn call_site() -> Span {
|
|
Span::_new(imp::Span::call_site())
|
|
}
|
|
|
|
/// The span located at the invocation of the procedural macro, but with
|
|
/// local variables, labels, and `$crate` resolved at the definition site
|
|
/// of the macro. This is the same hygiene behavior as `macro_rules`.
|
|
///
|
|
/// This function requires Rust 1.45 or later.
|
|
#[cfg(hygiene)]
|
|
pub fn mixed_site() -> Span {
|
|
Span::_new(imp::Span::mixed_site())
|
|
}
|
|
|
|
/// A span that resolves at the macro definition site.
|
|
///
|
|
/// This method is semver exempt and not exposed by default.
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(procmacro2_semver_exempt)))]
|
|
pub fn def_site() -> Span {
|
|
Span::_new(imp::Span::def_site())
|
|
}
|
|
|
|
/// Creates a new span with the same line/column information as `self` but
|
|
/// that resolves symbols as though it were at `other`.
|
|
pub fn resolved_at(&self, other: Span) -> Span {
|
|
Span::_new(self.inner.resolved_at(other.inner))
|
|
}
|
|
|
|
/// Creates a new span with the same name resolution behavior as `self` but
|
|
/// with the line/column information of `other`.
|
|
pub fn located_at(&self, other: Span) -> Span {
|
|
Span::_new(self.inner.located_at(other.inner))
|
|
}
|
|
|
|
/// Convert `proc_macro2::Span` to `proc_macro::Span`.
|
|
///
|
|
/// This method is available when building with a nightly compiler, or when
|
|
/// building with rustc 1.29+ *without* semver exempt features.
|
|
///
|
|
/// # Panics
|
|
///
|
|
/// Panics if called from outside of a procedural macro. Unlike
|
|
/// `proc_macro2::Span`, the `proc_macro::Span` type can only exist within
|
|
/// the context of a procedural macro invocation.
|
|
#[cfg(wrap_proc_macro)]
|
|
pub fn unwrap(self) -> proc_macro::Span {
|
|
self.inner.unwrap()
|
|
}
|
|
|
|
// Soft deprecated. Please use Span::unwrap.
|
|
#[cfg(wrap_proc_macro)]
|
|
#[doc(hidden)]
|
|
pub fn unstable(self) -> proc_macro::Span {
|
|
self.unwrap()
|
|
}
|
|
|
|
/// The original source file into which this span points.
|
|
///
|
|
/// This method is semver exempt and not exposed by default.
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(procmacro2_semver_exempt)))]
|
|
pub fn source_file(&self) -> SourceFile {
|
|
SourceFile::_new(self.inner.source_file())
|
|
}
|
|
|
|
/// Get the starting line/column in the source file for this span.
|
|
///
|
|
/// This method requires the `"span-locations"` feature to be enabled.
|
|
///
|
|
/// When executing in a procedural macro context, the returned line/column
|
|
/// are only meaningful if compiled with a nightly toolchain. The stable
|
|
/// toolchain does not have this information available. When executing
|
|
/// outside of a procedural macro, such as main.rs or build.rs, the
|
|
/// line/column are always meaningful regardless of toolchain.
|
|
#[cfg(span_locations)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(feature = "span-locations")))]
|
|
pub fn start(&self) -> LineColumn {
|
|
let imp::LineColumn { line, column } = self.inner.start();
|
|
LineColumn { line, column }
|
|
}
|
|
|
|
/// Get the ending line/column in the source file for this span.
|
|
///
|
|
/// This method requires the `"span-locations"` feature to be enabled.
|
|
///
|
|
/// When executing in a procedural macro context, the returned line/column
|
|
/// are only meaningful if compiled with a nightly toolchain. The stable
|
|
/// toolchain does not have this information available. When executing
|
|
/// outside of a procedural macro, such as main.rs or build.rs, the
|
|
/// line/column are always meaningful regardless of toolchain.
|
|
#[cfg(span_locations)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(feature = "span-locations")))]
|
|
pub fn end(&self) -> LineColumn {
|
|
let imp::LineColumn { line, column } = self.inner.end();
|
|
LineColumn { line, column }
|
|
}
|
|
|
|
/// Create a new span encompassing `self` and `other`.
|
|
///
|
|
/// Returns `None` if `self` and `other` are from different files.
|
|
///
|
|
/// Warning: the underlying [`proc_macro::Span::join`] method is
|
|
/// nightly-only. When called from within a procedural macro not using a
|
|
/// nightly compiler, this method will always return `None`.
|
|
///
|
|
/// [`proc_macro::Span::join`]: https://doc.rust-lang.org/proc_macro/struct.Span.html#method.join
|
|
pub fn join(&self, other: Span) -> Option<Span> {
|
|
self.inner.join(other.inner).map(Span::_new)
|
|
}
|
|
|
|
/// Compares two spans to see if they're equal.
|
|
///
|
|
/// This method is semver exempt and not exposed by default.
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(procmacro2_semver_exempt)))]
|
|
pub fn eq(&self, other: &Span) -> bool {
|
|
self.inner.eq(&other.inner)
|
|
}
|
|
}
|
|
|
|
/// Prints a span in a form convenient for debugging.
|
|
impl Debug for Span {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
/// A single token or a delimited sequence of token trees (e.g. `[1, (), ..]`).
|
|
#[derive(Clone)]
|
|
pub enum TokenTree {
|
|
/// A token stream surrounded by bracket delimiters.
|
|
Group(Group),
|
|
/// An identifier.
|
|
Ident(Ident),
|
|
/// A single punctuation character (`+`, `,`, `$`, etc.).
|
|
Punct(Punct),
|
|
/// A literal character (`'a'`), string (`"hello"`), number (`2.3`), etc.
|
|
Literal(Literal),
|
|
}
|
|
|
|
impl TokenTree {
|
|
/// Returns the span of this tree, delegating to the `span` method of
|
|
/// the contained token or a delimited stream.
|
|
pub fn span(&self) -> Span {
|
|
match self {
|
|
TokenTree::Group(t) => t.span(),
|
|
TokenTree::Ident(t) => t.span(),
|
|
TokenTree::Punct(t) => t.span(),
|
|
TokenTree::Literal(t) => t.span(),
|
|
}
|
|
}
|
|
|
|
/// Configures the span for *only this token*.
|
|
///
|
|
/// Note that if this token is a `Group` then this method will not configure
|
|
/// the span of each of the internal tokens, this will simply delegate to
|
|
/// the `set_span` method of each variant.
|
|
pub fn set_span(&mut self, span: Span) {
|
|
match self {
|
|
TokenTree::Group(t) => t.set_span(span),
|
|
TokenTree::Ident(t) => t.set_span(span),
|
|
TokenTree::Punct(t) => t.set_span(span),
|
|
TokenTree::Literal(t) => t.set_span(span),
|
|
}
|
|
}
|
|
}
|
|
|
|
impl From<Group> for TokenTree {
|
|
fn from(g: Group) -> TokenTree {
|
|
TokenTree::Group(g)
|
|
}
|
|
}
|
|
|
|
impl From<Ident> for TokenTree {
|
|
fn from(g: Ident) -> TokenTree {
|
|
TokenTree::Ident(g)
|
|
}
|
|
}
|
|
|
|
impl From<Punct> for TokenTree {
|
|
fn from(g: Punct) -> TokenTree {
|
|
TokenTree::Punct(g)
|
|
}
|
|
}
|
|
|
|
impl From<Literal> for TokenTree {
|
|
fn from(g: Literal) -> TokenTree {
|
|
TokenTree::Literal(g)
|
|
}
|
|
}
|
|
|
|
/// Prints the token tree as a string that is supposed to be losslessly
|
|
/// convertible back into the same token tree (modulo spans), except for
|
|
/// possibly `TokenTree::Group`s with `Delimiter::None` delimiters and negative
|
|
/// numeric literals.
|
|
impl Display for TokenTree {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
match self {
|
|
TokenTree::Group(t) => Display::fmt(t, f),
|
|
TokenTree::Ident(t) => Display::fmt(t, f),
|
|
TokenTree::Punct(t) => Display::fmt(t, f),
|
|
TokenTree::Literal(t) => Display::fmt(t, f),
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Prints token tree in a form convenient for debugging.
|
|
impl Debug for TokenTree {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
// Each of these has the name in the struct type in the derived debug,
|
|
// so don't bother with an extra layer of indirection
|
|
match self {
|
|
TokenTree::Group(t) => Debug::fmt(t, f),
|
|
TokenTree::Ident(t) => {
|
|
let mut debug = f.debug_struct("Ident");
|
|
debug.field("sym", &format_args!("{}", t));
|
|
imp::debug_span_field_if_nontrivial(&mut debug, t.span().inner);
|
|
debug.finish()
|
|
}
|
|
TokenTree::Punct(t) => Debug::fmt(t, f),
|
|
TokenTree::Literal(t) => Debug::fmt(t, f),
|
|
}
|
|
}
|
|
}
|
|
|
|
/// A delimited token stream.
|
|
///
|
|
/// A `Group` internally contains a `TokenStream` which is surrounded by
|
|
/// `Delimiter`s.
|
|
#[derive(Clone)]
|
|
pub struct Group {
|
|
inner: imp::Group,
|
|
}
|
|
|
|
/// Describes how a sequence of token trees is delimited.
|
|
#[derive(Copy, Clone, Debug, Eq, PartialEq)]
|
|
pub enum Delimiter {
|
|
/// `( ... )`
|
|
Parenthesis,
|
|
/// `{ ... }`
|
|
Brace,
|
|
/// `[ ... ]`
|
|
Bracket,
|
|
/// `Ø ... Ø`
|
|
///
|
|
/// An implicit delimiter, that may, for example, appear around tokens
|
|
/// coming from a "macro variable" `$var`. It is important to preserve
|
|
/// operator priorities in cases like `$var * 3` where `$var` is `1 + 2`.
|
|
/// Implicit delimiters may not survive roundtrip of a token stream through
|
|
/// a string.
|
|
None,
|
|
}
|
|
|
|
impl Group {
|
|
fn _new(inner: imp::Group) -> Self {
|
|
Group { inner }
|
|
}
|
|
|
|
fn _new_stable(inner: fallback::Group) -> Self {
|
|
Group {
|
|
inner: inner.into(),
|
|
}
|
|
}
|
|
|
|
/// Creates a new `Group` with the given delimiter and token stream.
|
|
///
|
|
/// This constructor will set the span for this group to
|
|
/// `Span::call_site()`. To change the span you can use the `set_span`
|
|
/// method below.
|
|
pub fn new(delimiter: Delimiter, stream: TokenStream) -> Group {
|
|
Group {
|
|
inner: imp::Group::new(delimiter, stream.inner),
|
|
}
|
|
}
|
|
|
|
/// Returns the delimiter of this `Group`
|
|
pub fn delimiter(&self) -> Delimiter {
|
|
self.inner.delimiter()
|
|
}
|
|
|
|
/// Returns the `TokenStream` of tokens that are delimited in this `Group`.
|
|
///
|
|
/// Note that the returned token stream does not include the delimiter
|
|
/// returned above.
|
|
pub fn stream(&self) -> TokenStream {
|
|
TokenStream::_new(self.inner.stream())
|
|
}
|
|
|
|
/// Returns the span for the delimiters of this token stream, spanning the
|
|
/// entire `Group`.
|
|
///
|
|
/// ```text
|
|
/// pub fn span(&self) -> Span {
|
|
/// ^^^^^^^
|
|
/// ```
|
|
pub fn span(&self) -> Span {
|
|
Span::_new(self.inner.span())
|
|
}
|
|
|
|
/// Returns the span pointing to the opening delimiter of this group.
|
|
///
|
|
/// ```text
|
|
/// pub fn span_open(&self) -> Span {
|
|
/// ^
|
|
/// ```
|
|
pub fn span_open(&self) -> Span {
|
|
Span::_new(self.inner.span_open())
|
|
}
|
|
|
|
/// Returns the span pointing to the closing delimiter of this group.
|
|
///
|
|
/// ```text
|
|
/// pub fn span_close(&self) -> Span {
|
|
/// ^
|
|
/// ```
|
|
pub fn span_close(&self) -> Span {
|
|
Span::_new(self.inner.span_close())
|
|
}
|
|
|
|
/// Configures the span for this `Group`'s delimiters, but not its internal
|
|
/// tokens.
|
|
///
|
|
/// This method will **not** set the span of all the internal tokens spanned
|
|
/// by this group, but rather it will only set the span of the delimiter
|
|
/// tokens at the level of the `Group`.
|
|
pub fn set_span(&mut self, span: Span) {
|
|
self.inner.set_span(span.inner)
|
|
}
|
|
}
|
|
|
|
/// Prints the group as a string that should be losslessly convertible back
|
|
/// into the same group (modulo spans), except for possibly `TokenTree::Group`s
|
|
/// with `Delimiter::None` delimiters.
|
|
impl Display for Group {
|
|
fn fmt(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
|
|
Display::fmt(&self.inner, formatter)
|
|
}
|
|
}
|
|
|
|
impl Debug for Group {
|
|
fn fmt(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, formatter)
|
|
}
|
|
}
|
|
|
|
/// A `Punct` is a single punctuation character like `+`, `-` or `#`.
|
|
///
|
|
/// Multicharacter operators like `+=` are represented as two instances of
|
|
/// `Punct` with different forms of `Spacing` returned.
|
|
#[derive(Clone)]
|
|
pub struct Punct {
|
|
ch: char,
|
|
spacing: Spacing,
|
|
span: Span,
|
|
}
|
|
|
|
/// Whether a `Punct` is followed immediately by another `Punct` or followed by
|
|
/// another token or whitespace.
|
|
#[derive(Copy, Clone, Debug, Eq, PartialEq)]
|
|
pub enum Spacing {
|
|
/// E.g. `+` is `Alone` in `+ =`, `+ident` or `+()`.
|
|
Alone,
|
|
/// E.g. `+` is `Joint` in `+=` or `'` is `Joint` in `'#`.
|
|
///
|
|
/// Additionally, single quote `'` can join with identifiers to form
|
|
/// lifetimes `'ident`.
|
|
Joint,
|
|
}
|
|
|
|
impl Punct {
|
|
/// Creates a new `Punct` from the given character and spacing.
|
|
///
|
|
/// The `ch` argument must be a valid punctuation character permitted by the
|
|
/// language, otherwise the function will panic.
|
|
///
|
|
/// The returned `Punct` will have the default span of `Span::call_site()`
|
|
/// which can be further configured with the `set_span` method below.
|
|
pub fn new(ch: char, spacing: Spacing) -> Punct {
|
|
Punct {
|
|
ch,
|
|
spacing,
|
|
span: Span::call_site(),
|
|
}
|
|
}
|
|
|
|
/// Returns the value of this punctuation character as `char`.
|
|
pub fn as_char(&self) -> char {
|
|
self.ch
|
|
}
|
|
|
|
/// Returns the spacing of this punctuation character, indicating whether
|
|
/// it's immediately followed by another `Punct` in the token stream, so
|
|
/// they can potentially be combined into a multicharacter operator
|
|
/// (`Joint`), or it's followed by some other token or whitespace (`Alone`)
|
|
/// so the operator has certainly ended.
|
|
pub fn spacing(&self) -> Spacing {
|
|
self.spacing
|
|
}
|
|
|
|
/// Returns the span for this punctuation character.
|
|
pub fn span(&self) -> Span {
|
|
self.span
|
|
}
|
|
|
|
/// Configure the span for this punctuation character.
|
|
pub fn set_span(&mut self, span: Span) {
|
|
self.span = span;
|
|
}
|
|
}
|
|
|
|
/// Prints the punctuation character as a string that should be losslessly
|
|
/// convertible back into the same character.
|
|
impl Display for Punct {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Display::fmt(&self.ch, f)
|
|
}
|
|
}
|
|
|
|
impl Debug for Punct {
|
|
fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result {
|
|
let mut debug = fmt.debug_struct("Punct");
|
|
debug.field("char", &self.ch);
|
|
debug.field("spacing", &self.spacing);
|
|
imp::debug_span_field_if_nontrivial(&mut debug, self.span.inner);
|
|
debug.finish()
|
|
}
|
|
}
|
|
|
|
/// A word of Rust code, which may be a keyword or legal variable name.
|
|
///
|
|
/// An identifier consists of at least one Unicode code point, the first of
|
|
/// which has the XID_Start property and the rest of which have the XID_Continue
|
|
/// property.
|
|
///
|
|
/// - The empty string is not an identifier. Use `Option<Ident>`.
|
|
/// - A lifetime is not an identifier. Use `syn::Lifetime` instead.
|
|
///
|
|
/// An identifier constructed with `Ident::new` is permitted to be a Rust
|
|
/// keyword, though parsing one through its [`Parse`] implementation rejects
|
|
/// Rust keywords. Use `input.call(Ident::parse_any)` when parsing to match the
|
|
/// behaviour of `Ident::new`.
|
|
///
|
|
/// [`Parse`]: https://docs.rs/syn/1.0/syn/parse/trait.Parse.html
|
|
///
|
|
/// # Examples
|
|
///
|
|
/// A new ident can be created from a string using the `Ident::new` function.
|
|
/// A span must be provided explicitly which governs the name resolution
|
|
/// behavior of the resulting identifier.
|
|
///
|
|
/// ```
|
|
/// use proc_macro2::{Ident, Span};
|
|
///
|
|
/// fn main() {
|
|
/// let call_ident = Ident::new("calligraphy", Span::call_site());
|
|
///
|
|
/// println!("{}", call_ident);
|
|
/// }
|
|
/// ```
|
|
///
|
|
/// An ident can be interpolated into a token stream using the `quote!` macro.
|
|
///
|
|
/// ```
|
|
/// use proc_macro2::{Ident, Span};
|
|
/// use quote::quote;
|
|
///
|
|
/// fn main() {
|
|
/// let ident = Ident::new("demo", Span::call_site());
|
|
///
|
|
/// // Create a variable binding whose name is this ident.
|
|
/// let expanded = quote! { let #ident = 10; };
|
|
///
|
|
/// // Create a variable binding with a slightly different name.
|
|
/// let temp_ident = Ident::new(&format!("new_{}", ident), Span::call_site());
|
|
/// let expanded = quote! { let #temp_ident = 10; };
|
|
/// }
|
|
/// ```
|
|
///
|
|
/// A string representation of the ident is available through the `to_string()`
|
|
/// method.
|
|
///
|
|
/// ```
|
|
/// # use proc_macro2::{Ident, Span};
|
|
/// #
|
|
/// # let ident = Ident::new("another_identifier", Span::call_site());
|
|
/// #
|
|
/// // Examine the ident as a string.
|
|
/// let ident_string = ident.to_string();
|
|
/// if ident_string.len() > 60 {
|
|
/// println!("Very long identifier: {}", ident_string)
|
|
/// }
|
|
/// ```
|
|
#[derive(Clone)]
|
|
pub struct Ident {
|
|
inner: imp::Ident,
|
|
_marker: Marker,
|
|
}
|
|
|
|
impl Ident {
|
|
fn _new(inner: imp::Ident) -> Ident {
|
|
Ident {
|
|
inner,
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
/// Creates a new `Ident` with the given `string` as well as the specified
|
|
/// `span`.
|
|
///
|
|
/// The `string` argument must be a valid identifier permitted by the
|
|
/// language, otherwise the function will panic.
|
|
///
|
|
/// Note that `span`, currently in rustc, configures the hygiene information
|
|
/// for this identifier.
|
|
///
|
|
/// As of this time `Span::call_site()` explicitly opts-in to "call-site"
|
|
/// hygiene meaning that identifiers created with this span will be resolved
|
|
/// as if they were written directly at the location of the macro call, and
|
|
/// other code at the macro call site will be able to refer to them as well.
|
|
///
|
|
/// Later spans like `Span::def_site()` will allow to opt-in to
|
|
/// "definition-site" hygiene meaning that identifiers created with this
|
|
/// span will be resolved at the location of the macro definition and other
|
|
/// code at the macro call site will not be able to refer to them.
|
|
///
|
|
/// Due to the current importance of hygiene this constructor, unlike other
|
|
/// tokens, requires a `Span` to be specified at construction.
|
|
///
|
|
/// # Panics
|
|
///
|
|
/// Panics if the input string is neither a keyword nor a legal variable
|
|
/// name. If you are not sure whether the string contains an identifier and
|
|
/// need to handle an error case, use
|
|
/// <a href="https://docs.rs/syn/1.0/syn/fn.parse_str.html"><code
|
|
/// style="padding-right:0;">syn::parse_str</code></a><code
|
|
/// style="padding-left:0;">::<Ident></code>
|
|
/// rather than `Ident::new`.
|
|
pub fn new(string: &str, span: Span) -> Ident {
|
|
Ident::_new(imp::Ident::new(string, span.inner))
|
|
}
|
|
|
|
/// Same as `Ident::new`, but creates a raw identifier (`r#ident`).
|
|
///
|
|
/// This method is semver exempt and not exposed by default.
|
|
#[cfg(procmacro2_semver_exempt)]
|
|
#[cfg_attr(doc_cfg, doc(cfg(procmacro2_semver_exempt)))]
|
|
pub fn new_raw(string: &str, span: Span) -> Ident {
|
|
Ident::_new_raw(string, span)
|
|
}
|
|
|
|
fn _new_raw(string: &str, span: Span) -> Ident {
|
|
Ident::_new(imp::Ident::new_raw(string, span.inner))
|
|
}
|
|
|
|
/// Returns the span of this `Ident`.
|
|
pub fn span(&self) -> Span {
|
|
Span::_new(self.inner.span())
|
|
}
|
|
|
|
/// Configures the span of this `Ident`, possibly changing its hygiene
|
|
/// context.
|
|
pub fn set_span(&mut self, span: Span) {
|
|
self.inner.set_span(span.inner);
|
|
}
|
|
}
|
|
|
|
impl PartialEq for Ident {
|
|
fn eq(&self, other: &Ident) -> bool {
|
|
self.inner == other.inner
|
|
}
|
|
}
|
|
|
|
impl<T> PartialEq<T> for Ident
|
|
where
|
|
T: ?Sized + AsRef<str>,
|
|
{
|
|
fn eq(&self, other: &T) -> bool {
|
|
self.inner == other
|
|
}
|
|
}
|
|
|
|
impl Eq for Ident {}
|
|
|
|
impl PartialOrd for Ident {
|
|
fn partial_cmp(&self, other: &Ident) -> Option<Ordering> {
|
|
Some(self.cmp(other))
|
|
}
|
|
}
|
|
|
|
impl Ord for Ident {
|
|
fn cmp(&self, other: &Ident) -> Ordering {
|
|
self.to_string().cmp(&other.to_string())
|
|
}
|
|
}
|
|
|
|
impl Hash for Ident {
|
|
fn hash<H: Hasher>(&self, hasher: &mut H) {
|
|
self.to_string().hash(hasher)
|
|
}
|
|
}
|
|
|
|
/// Prints the identifier as a string that should be losslessly convertible back
|
|
/// into the same identifier.
|
|
impl Display for Ident {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Display::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
impl Debug for Ident {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
/// A literal string (`"hello"`), byte string (`b"hello"`), character (`'a'`),
|
|
/// byte character (`b'a'`), an integer or floating point number with or without
|
|
/// a suffix (`1`, `1u8`, `2.3`, `2.3f32`).
|
|
///
|
|
/// Boolean literals like `true` and `false` do not belong here, they are
|
|
/// `Ident`s.
|
|
#[derive(Clone)]
|
|
pub struct Literal {
|
|
inner: imp::Literal,
|
|
_marker: Marker,
|
|
}
|
|
|
|
macro_rules! suffixed_int_literals {
|
|
($($name:ident => $kind:ident,)*) => ($(
|
|
/// Creates a new suffixed integer literal with the specified value.
|
|
///
|
|
/// This function will create an integer like `1u32` where the integer
|
|
/// value specified is the first part of the token and the integral is
|
|
/// also suffixed at the end. Literals created from negative numbers may
|
|
/// not survive roundtrips through `TokenStream` or strings and may be
|
|
/// broken into two tokens (`-` and positive literal).
|
|
///
|
|
/// Literals created through this method have the `Span::call_site()`
|
|
/// span by default, which can be configured with the `set_span` method
|
|
/// below.
|
|
pub fn $name(n: $kind) -> Literal {
|
|
Literal::_new(imp::Literal::$name(n))
|
|
}
|
|
)*)
|
|
}
|
|
|
|
macro_rules! unsuffixed_int_literals {
|
|
($($name:ident => $kind:ident,)*) => ($(
|
|
/// Creates a new unsuffixed integer literal with the specified value.
|
|
///
|
|
/// This function will create an integer like `1` where the integer
|
|
/// value specified is the first part of the token. No suffix is
|
|
/// specified on this token, meaning that invocations like
|
|
/// `Literal::i8_unsuffixed(1)` are equivalent to
|
|
/// `Literal::u32_unsuffixed(1)`. Literals created from negative numbers
|
|
/// may not survive roundtrips through `TokenStream` or strings and may
|
|
/// be broken into two tokens (`-` and positive literal).
|
|
///
|
|
/// Literals created through this method have the `Span::call_site()`
|
|
/// span by default, which can be configured with the `set_span` method
|
|
/// below.
|
|
pub fn $name(n: $kind) -> Literal {
|
|
Literal::_new(imp::Literal::$name(n))
|
|
}
|
|
)*)
|
|
}
|
|
|
|
impl Literal {
|
|
fn _new(inner: imp::Literal) -> Literal {
|
|
Literal {
|
|
inner,
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
fn _new_stable(inner: fallback::Literal) -> Literal {
|
|
Literal {
|
|
inner: inner.into(),
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
|
|
suffixed_int_literals! {
|
|
u8_suffixed => u8,
|
|
u16_suffixed => u16,
|
|
u32_suffixed => u32,
|
|
u64_suffixed => u64,
|
|
u128_suffixed => u128,
|
|
usize_suffixed => usize,
|
|
i8_suffixed => i8,
|
|
i16_suffixed => i16,
|
|
i32_suffixed => i32,
|
|
i64_suffixed => i64,
|
|
i128_suffixed => i128,
|
|
isize_suffixed => isize,
|
|
}
|
|
|
|
unsuffixed_int_literals! {
|
|
u8_unsuffixed => u8,
|
|
u16_unsuffixed => u16,
|
|
u32_unsuffixed => u32,
|
|
u64_unsuffixed => u64,
|
|
u128_unsuffixed => u128,
|
|
usize_unsuffixed => usize,
|
|
i8_unsuffixed => i8,
|
|
i16_unsuffixed => i16,
|
|
i32_unsuffixed => i32,
|
|
i64_unsuffixed => i64,
|
|
i128_unsuffixed => i128,
|
|
isize_unsuffixed => isize,
|
|
}
|
|
|
|
/// Creates a new unsuffixed floating-point literal.
|
|
///
|
|
/// This constructor is similar to those like `Literal::i8_unsuffixed` where
|
|
/// the float's value is emitted directly into the token but no suffix is
|
|
/// used, so it may be inferred to be a `f64` later in the compiler.
|
|
/// Literals created from negative numbers may not survive rountrips through
|
|
/// `TokenStream` or strings and may be broken into two tokens (`-` and
|
|
/// positive literal).
|
|
///
|
|
/// # Panics
|
|
///
|
|
/// This function requires that the specified float is finite, for example
|
|
/// if it is infinity or NaN this function will panic.
|
|
pub fn f64_unsuffixed(f: f64) -> Literal {
|
|
assert!(f.is_finite());
|
|
Literal::_new(imp::Literal::f64_unsuffixed(f))
|
|
}
|
|
|
|
/// Creates a new suffixed floating-point literal.
|
|
///
|
|
/// This constructor will create a literal like `1.0f64` where the value
|
|
/// specified is the preceding part of the token and `f64` is the suffix of
|
|
/// the token. This token will always be inferred to be an `f64` in the
|
|
/// compiler. Literals created from negative numbers may not survive
|
|
/// rountrips through `TokenStream` or strings and may be broken into two
|
|
/// tokens (`-` and positive literal).
|
|
///
|
|
/// # Panics
|
|
///
|
|
/// This function requires that the specified float is finite, for example
|
|
/// if it is infinity or NaN this function will panic.
|
|
pub fn f64_suffixed(f: f64) -> Literal {
|
|
assert!(f.is_finite());
|
|
Literal::_new(imp::Literal::f64_suffixed(f))
|
|
}
|
|
|
|
/// Creates a new unsuffixed floating-point literal.
|
|
///
|
|
/// This constructor is similar to those like `Literal::i8_unsuffixed` where
|
|
/// the float's value is emitted directly into the token but no suffix is
|
|
/// used, so it may be inferred to be a `f64` later in the compiler.
|
|
/// Literals created from negative numbers may not survive rountrips through
|
|
/// `TokenStream` or strings and may be broken into two tokens (`-` and
|
|
/// positive literal).
|
|
///
|
|
/// # Panics
|
|
///
|
|
/// This function requires that the specified float is finite, for example
|
|
/// if it is infinity or NaN this function will panic.
|
|
pub fn f32_unsuffixed(f: f32) -> Literal {
|
|
assert!(f.is_finite());
|
|
Literal::_new(imp::Literal::f32_unsuffixed(f))
|
|
}
|
|
|
|
/// Creates a new suffixed floating-point literal.
|
|
///
|
|
/// This constructor will create a literal like `1.0f32` where the value
|
|
/// specified is the preceding part of the token and `f32` is the suffix of
|
|
/// the token. This token will always be inferred to be an `f32` in the
|
|
/// compiler. Literals created from negative numbers may not survive
|
|
/// rountrips through `TokenStream` or strings and may be broken into two
|
|
/// tokens (`-` and positive literal).
|
|
///
|
|
/// # Panics
|
|
///
|
|
/// This function requires that the specified float is finite, for example
|
|
/// if it is infinity or NaN this function will panic.
|
|
pub fn f32_suffixed(f: f32) -> Literal {
|
|
assert!(f.is_finite());
|
|
Literal::_new(imp::Literal::f32_suffixed(f))
|
|
}
|
|
|
|
/// String literal.
|
|
pub fn string(string: &str) -> Literal {
|
|
Literal::_new(imp::Literal::string(string))
|
|
}
|
|
|
|
/// Character literal.
|
|
pub fn character(ch: char) -> Literal {
|
|
Literal::_new(imp::Literal::character(ch))
|
|
}
|
|
|
|
/// Byte string literal.
|
|
pub fn byte_string(s: &[u8]) -> Literal {
|
|
Literal::_new(imp::Literal::byte_string(s))
|
|
}
|
|
|
|
/// Returns the span encompassing this literal.
|
|
pub fn span(&self) -> Span {
|
|
Span::_new(self.inner.span())
|
|
}
|
|
|
|
/// Configures the span associated for this literal.
|
|
pub fn set_span(&mut self, span: Span) {
|
|
self.inner.set_span(span.inner);
|
|
}
|
|
|
|
/// Returns a `Span` that is a subset of `self.span()` containing only
|
|
/// the source bytes in range `range`. Returns `None` if the would-be
|
|
/// trimmed span is outside the bounds of `self`.
|
|
///
|
|
/// Warning: the underlying [`proc_macro::Literal::subspan`] method is
|
|
/// nightly-only. When called from within a procedural macro not using a
|
|
/// nightly compiler, this method will always return `None`.
|
|
///
|
|
/// [`proc_macro::Literal::subspan`]: https://doc.rust-lang.org/proc_macro/struct.Literal.html#method.subspan
|
|
pub fn subspan<R: RangeBounds<usize>>(&self, range: R) -> Option<Span> {
|
|
self.inner.subspan(range).map(Span::_new)
|
|
}
|
|
}
|
|
|
|
impl FromStr for Literal {
|
|
type Err = LexError;
|
|
|
|
fn from_str(repr: &str) -> Result<Self, LexError> {
|
|
repr.parse().map(Literal::_new).map_err(|inner| LexError {
|
|
inner,
|
|
_marker: Marker,
|
|
})
|
|
}
|
|
}
|
|
|
|
impl Debug for Literal {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
impl Display for Literal {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Display::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
/// Public implementation details for the `TokenStream` type, such as iterators.
|
|
pub mod token_stream {
|
|
use crate::marker::Marker;
|
|
use crate::{imp, TokenTree};
|
|
use std::fmt::{self, Debug};
|
|
|
|
pub use crate::TokenStream;
|
|
|
|
/// An iterator over `TokenStream`'s `TokenTree`s.
|
|
///
|
|
/// The iteration is "shallow", e.g. the iterator doesn't recurse into
|
|
/// delimited groups, and returns whole groups as token trees.
|
|
#[derive(Clone)]
|
|
pub struct IntoIter {
|
|
inner: imp::TokenTreeIter,
|
|
_marker: Marker,
|
|
}
|
|
|
|
impl Iterator for IntoIter {
|
|
type Item = TokenTree;
|
|
|
|
fn next(&mut self) -> Option<TokenTree> {
|
|
self.inner.next()
|
|
}
|
|
}
|
|
|
|
impl Debug for IntoIter {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
Debug::fmt(&self.inner, f)
|
|
}
|
|
}
|
|
|
|
impl IntoIterator for TokenStream {
|
|
type Item = TokenTree;
|
|
type IntoIter = IntoIter;
|
|
|
|
fn into_iter(self) -> IntoIter {
|
|
IntoIter {
|
|
inner: self.inner.into_iter(),
|
|
_marker: Marker,
|
|
}
|
|
}
|
|
}
|
|
}
|