如何编写生成模块的 Rust 编译器插件?

How to write a Rust compiler plugin that generates a module?

我正在编写一个扩展

的 Rust 编译器插件
choose! {
    test_a
    test_b
}

#[cfg(feature = "a")]
mod test_a;
#[cfg(feature = "b")]
mod test_b;

差不多大功告成了,但是最后展开的代码中模块什么都没有。我想原因是跨度没有覆盖模块文件。

use syntax::ast;
use syntax::ptr::P;
use syntax::codemap;
use syntax::parse::token;
use syntax::tokenstream::TokenTree;
use syntax::ext::base::{ExtCtxt, MacResult, DummyResult, MacEager};
use syntax::ext::build::AstBuilder;
use syntax_pos::Span;
use rustc_plugin::Registry;
use syntax::util::small_vector::SmallVector;

// Ideally, it will expand
//
// ```rust
// choose! {
//   test_a
//   test_b
// }
// ```
// to
// ```rust
// #[cfg(feature = "a")]
// mod test_a;
// #[cfg(feature = "b")]
// mod test_b;
// ```
//
// but the modules contain nothing in the expanded code at present

fn choose(cx: &mut ExtCtxt, sp: Span, args: &[TokenTree]) -> Box<MacResult + 'static> {
    let mut test_mods: SmallVector<P<ast::Item>> = SmallVector::many(vec![]);
    for arg in args {
        let mut attrs = vec![];
        let text = match arg {
            &TokenTree::Token(_, token::Ident(s)) => s.to_string(),
            _ => {
                return DummyResult::any(sp);
            }
        };
        let cfg_str = token::InternedString::new("cfg");
        let feat_str = token::InternedString::new("feature");
        attrs.push(cx.attribute(sp,
                                cx.meta_list(sp,
                                             cfg_str,
                                             vec![cx.meta_name_value(sp,
                                                                     feat_str,
                                                                     ast::LitKind::Str(token::intern_and_get_ident(text.trim_left_matches("test_")), ast::StrStyle::Cooked))])));
        test_mods.push(P(ast::Item {
            ident: cx.ident_of(text.as_str()),
            attrs: attrs,
            id: ast::DUMMY_NODE_ID,
            node: ast::ItemKind::Mod(
                // === How to include the specified module file here? ===
                ast::Mod {
                    inner: codemap::DUMMY_SP,
                    items: vec![],
                }
            ),
            vis: ast::Visibility::Inherited,
            span: sp,
        }))
    }

    MacEager::items(test_mods)
}

#[plugin_registrar]
pub fn plugin_registrar(reg: &mut Registry) {
    reg.register_macro("choose", choose);
}

(Gist)

2016-08-25更新:使用libsyntax::parse::new_parser_from_source_str避免手动设置模块路径。new_parser_from_source_str只会在CWD定位模块,这是意想不到的.

正如@Francis 所指出的,模块文件的真实路径可能类似于 foo/mod.rs, 我发现了一个名为 new_parser_from_source_str 的函数,它能够创建来自源字符串的新解析器,在 libsyntax::parse 中,所以我决定让编译器为我处理这种情况 所以我必须手动处理这种情况。更新后的代码:

fn choose(cx: &mut ExtCtxt, sp: Span, args: &[TokenTree]) -> Box<MacResult + 'static> {
    let mut test_mods = SmallVector::zero();
    let cfg_str = intern("cfg");
    let ftre_str = intern("feature");
    for arg in args {
        let mut attrs = vec![];
        let mod_name = match arg {
            &TokenTree::Token(_, token::Ident(s)) => s.to_string(),
            _ => {
                return DummyResult::any(sp);
            }
        };
        attrs.push(cx.attribute(sp,
                                cx.meta_list(sp,
                                             // simply increase the reference counter
                                             cfg_str.clone(),
                                             vec![cx.meta_name_value(sp,
                                                                     ftre_str.clone(),
                                                                     ast::LitKind::Str(intern(mod_name.trim_left_matches("test_")), ast::StrStyle::Cooked))])));

        let mut mod_path = PathBuf::from(&cx.codemap().span_to_filename(sp));
        let dir = mod_path.parent().expect("no parent directory").to_owned();
        let default_path = dir.join(format!("{}.rs", mod_name.as_str()));
        let secondary_path = dir.join(format!("{}/mod.rs", mod_name.as_str()));
        match (default_path.exists(), secondary_path.exists()) {
            (false, false) => {
                cx.span_err(sp, &format!("file not found for module `{}`", mod_name.as_str()));
                return DummyResult::any(sp);
            }
            (true, true) => {
                cx.span_err(sp, &format!("file for module `{}` found at both {} and {}", mod_name.as_str(), default_path.display(), secondary_path.display()));
                return DummyResult::any(sp);
            }
            (true, false) => mod_path = default_path,
            (false, true) => mod_path = secondary_path,
        }

        test_mods.push(P(ast::Item {
            ident: cx.ident_of(mod_name.as_str()),
            attrs: attrs,
            id: ast::DUMMY_NODE_ID,
            node: ast::ItemKind::Mod(
                ast::Mod {
                    inner: sp,
                    items: expand_include(cx, sp, &mod_path),
                }
            ),
            vis: ast::Visibility::Inherited,
            span: sp,
        }))
    }

    MacEager::items(test_mods)
}

终于找到解决办法了! \o/

Rust 处理模块文件的过程类似于 include!。结果,我查看了宏include!的实现,可以找到here,并根据需要重写了它:

use ::std::path::Path;
use ::std::path::PathBuf;
use syntax::parse::{self, token};
use syntax::errors::FatalError;
macro_rules! panictry {
    ($e:expr) => ({
        match $e {
            Ok(e) => e,
            Err(mut e) => {
                e.emit();
                panic!(FatalError);
            }
        }
    })
}

pub fn expand_include<'cx>(cx: &'cx mut ExtCtxt, sp: Span, file: &Path) -> Vec<P<ast::Item>> {
    let mut p = parse::new_sub_parser_from_file(cx.parse_sess(), cx.cfg(), file, true, None, sp);
    let mut ret = vec![];
    while p.token != token::Eof {
        match panictry!(p.parse_item()) {
            Some(item) => ret.push(item),
            None => {
                panic!(p.diagnostic().span_fatal(p.span,
                                                 &format!("expected item, found `{}`", p.this_token_to_string())))
            }
        }
    }
    ret
}

要从模块文件中获取项目,我们必须找到真正的模块路径:

let mut mod_path = PathBuf::from(&cx.codemap().span_to_filename(sp));
mod_path.set_file_name(mod_name.as_str());
mod_path.set_extension("rs");

然后我们可以这样构造我们的模块节点:

P(ast::Item {
    ident: cx.ident_of(mod_name.as_str()),
    attrs: attrs,
    id: ast::DUMMY_NODE_ID,
    node: ast::ItemKind::Mod(ast::Mod {
        inner: sp,
        items: expand_include(cx, sp, &mod_path),
    }),
    vis: ast::Visibility::Inherited,
    span: sp,
})

综上,插件改写如下:

#![feature(plugin_registrar, rustc_private)]

extern crate syntax;
extern crate rustc_plugin;

use syntax::ast;
use syntax::ptr::P;
use syntax::codemap::Span;
use syntax::parse::{self, token};
use syntax::tokenstream::TokenTree;
use syntax::ext::base::{ExtCtxt, MacResult, DummyResult, MacEager};
use syntax::errors::FatalError;
use syntax::ext::build::AstBuilder;
use rustc_plugin::Registry;
use syntax::util::small_vector::SmallVector;

use ::std::path::Path;
use ::std::path::PathBuf;

macro_rules! panictry {
    ($e:expr) => ({
        match $e {
            Ok(e) => e,
            Err(mut e) => {
                e.emit();
                panic!(FatalError);
            }
        }
    })
}

pub fn expand_include<'cx>(cx: &'cx mut ExtCtxt, sp: Span, file: &Path) -> Vec<P<ast::Item>> {
    let mut p = parse::new_sub_parser_from_file(cx.parse_sess(), cx.cfg(), file, true, None, sp);
    let mut ret = vec![];
    while p.token != token::Eof {
        match panictry!(p.parse_item()) {
            Some(item) => ret.push(item),
            None => {
                panic!(p.diagnostic().span_fatal(p.span,
                                                 &format!("expected item, found `{}`", p.this_token_to_string())))
            }
        }
    }
    ret
}

fn intern(s: &str) -> token::InternedString {
    token::intern_and_get_ident(s)
}

fn choose(cx: &mut ExtCtxt, sp: Span, args: &[TokenTree]) -> Box<MacResult + 'static> {
    let mut test_mods = SmallVector::zero();
    let cfg_str = intern("cfg");
    let feat_str = intern("feature");
    for arg in args {
        let mut attrs = vec![];
        let mod_name = match arg {
            &TokenTree::Token(_, token::Ident(s)) => s.to_string(),
            _ => {
                return DummyResult::any(sp);
            }
        };
        attrs.push(cx.attribute(sp,
                                cx.meta_list(sp,
                                             // simply increase the reference counter
                                             cfg_str.clone(),
                                             vec![cx.meta_name_value(sp,
                                                                     feat_str.clone(),
                                                                     ast::LitKind::Str(intern(mod_name.trim_left_matches("test_")), ast::StrStyle::Cooked))])));

        let mut mod_path = PathBuf::from(&cx.codemap().span_to_filename(sp));
        mod_path.set_file_name(mod_name.as_str());
        mod_path.set_extension("rs");

        test_mods.push(P(ast::Item {
            ident: cx.ident_of(mod_name.as_str()),
            attrs: attrs,
            id: ast::DUMMY_NODE_ID,
            node: ast::ItemKind::Mod(
                ast::Mod {
                    inner: sp,
                    items: expand_include(cx, sp, &mod_path),
                }
            ),
            vis: ast::Visibility::Inherited,
            span: sp,
        }))
    }

    MacEager::items(test_mods)
}

#[plugin_registrar]
pub fn plugin_registrar(reg: &mut Registry) {
    reg.register_macro("choose", choose);
}