Add tool to read a Rust crate and generate C-compatible wrappers

In general, it maps:
 * Traits to a struct with a void* and a list of function pointers,
   emulating what the compiler will do for a dyn trait anyway,
 * Structs as a struct with a single opaque pointer to the
   underlying type and a flag to indicate ownership. While this is
   a bit less effecient than just a direct pointer, it neatly lets
   us expose in the public interface the concept of ownership by
   setting a flag in the generated struct.
 * Unit enums as enums with each type copied over and conversion
   functions,
 * Non-unit enums have each field converted back and forth with a
   type flag and a union across all the C-mapped fields.
This commit is contained in:
Matt Corallo 2020-05-12 14:49:29 -04:00
parent bd5f2427c3
commit 48e8678a9a
4 changed files with 3689 additions and 0 deletions

12
c-bindings-gen/Cargo.toml Normal file
View file

@ -0,0 +1,12 @@
[package]
name = "c-bindings-gen"
version = "0.0.1"
authors = ["Matt Corallo"]
edition = "2018"
[dependencies]
syn = { version = "1", features = ["full", "extra-traits"] }
proc-macro2 = "1"
# We're not in the workspace as we're just a binary code generator:
[workspace]

View file

@ -0,0 +1,373 @@
//! Printing logic for basic blocks of Rust-mapped code - parts of functions and declarations but
//! not the full mapping logic.
use std::collections::HashMap;
use std::fs::File;
use std::io::Write;
use proc_macro2::{TokenTree, Span};
use crate::types::*;
/// Writes out a C++ wrapper class for the given type, which contains various utilities to access
/// the underlying C-mapped type safely avoiding some common memory management issues by handling
/// resource-freeing and prevending accidental raw copies.
pub fn write_cpp_wrapper(cpp_header_file: &mut File, ty: &str, has_destructor: bool) {
writeln!(cpp_header_file, "class {} {{", ty).unwrap();
writeln!(cpp_header_file, "private:").unwrap();
writeln!(cpp_header_file, "\tLDK{} self;", ty).unwrap();
writeln!(cpp_header_file, "public:").unwrap();
writeln!(cpp_header_file, "\t{}(const {}&) = delete;", ty, ty).unwrap();
if has_destructor {
writeln!(cpp_header_file, "\t~{}() {{ {}_free(self); }}", ty, ty).unwrap();
}
writeln!(cpp_header_file, "\t{}({}&& o) : self(o.self) {{ memset(&o, 0, sizeof({})); }}", ty, ty, ty).unwrap();
writeln!(cpp_header_file, "\t{}(LDK{}&& m_self) : self(m_self) {{ memset(&m_self, 0, sizeof(LDK{})); }}", ty, ty, ty).unwrap();
writeln!(cpp_header_file, "\toperator LDK{}() {{ LDK{} res = self; memset(&self, 0, sizeof(LDK{})); return res; }}", ty, ty, ty).unwrap();
writeln!(cpp_header_file, "\tLDK{}* operator &() {{ return &self; }}", ty).unwrap();
writeln!(cpp_header_file, "\tLDK{}* operator ->() {{ return &self; }}", ty).unwrap();
writeln!(cpp_header_file, "\tconst LDK{}* operator &() const {{ return &self; }}", ty).unwrap();
writeln!(cpp_header_file, "\tconst LDK{}* operator ->() const {{ return &self; }}", ty).unwrap();
writeln!(cpp_header_file, "}};").unwrap();
}
/// Prints the docs from a given attribute list unless its tagged no export
pub fn writeln_docs<W: std::io::Write>(w: &mut W, attrs: &[syn::Attribute], prefix: &str) {
for attr in attrs.iter() {
let tokens_clone = attr.tokens.clone();
let mut token_iter = tokens_clone.into_iter();
if let Some(token) = token_iter.next() {
match token {
TokenTree::Punct(c) if c.as_char() == '=' => {
// syn gets '=' from '///' or '//!' as it is syntax for #[doc = ""]
},
TokenTree::Group(_) => continue, // eg #[derive()]
_ => unimplemented!(),
}
} else { continue; }
match attr.style {
syn::AttrStyle::Inner(_) => {
match token_iter.next().unwrap() {
TokenTree::Literal(lit) => {
// Drop the first and last chars from lit as they are always "
let doc = format!("{}", lit);
writeln!(w, "{}//!{}", prefix, &doc[1..doc.len() - 1]).unwrap();
},
_ => unimplemented!(),
}
},
syn::AttrStyle::Outer => {
match token_iter.next().unwrap() {
TokenTree::Literal(lit) => {
// Drop the first and last chars from lit as they are always "
let doc = format!("{}", lit);
writeln!(w, "{}///{}", prefix, &doc[1..doc.len() - 1]).unwrap();
},
_ => unimplemented!(),
}
},
}
}
}
/// Print the parameters in a method declaration, starting after the open parenthesis, through and
/// including the closing parenthesis and return value, but not including the open bracket or any
/// trailing semicolons.
///
/// Usable both for a function definition and declaration.
///
/// this_param is used when returning Self or accepting a self parameter, and should be the
/// concrete, mapped type.
pub fn write_method_params<W: std::io::Write>(w: &mut W, sig: &syn::Signature, associated_types: &HashMap<&syn::Ident, &syn::Ident>, this_param: &str, types: &mut TypeResolver, generics: Option<&GenericTypes>, self_ptr: bool, fn_decl: bool) {
if sig.constness.is_some() || sig.asyncness.is_some() || sig.unsafety.is_some() ||
sig.abi.is_some() || sig.variadic.is_some() {
unimplemented!();
}
if sig.generics.lt_token.is_some() {
for generic in sig.generics.params.iter() {
match generic {
syn::GenericParam::Type(_)|syn::GenericParam::Lifetime(_) => {
// We ignore these, if they're not on skipped args, we'll blow up
// later, and lifetimes we just hope the C client enforces.
},
_ => unimplemented!(),
}
}
}
let mut first_arg = true;
let mut num_unused = 0;
for inp in sig.inputs.iter() {
match inp {
syn::FnArg::Receiver(recv) => {
if !recv.attrs.is_empty() || recv.reference.is_none() { unimplemented!(); }
write!(w, "this_arg: {}{}",
match (self_ptr, recv.mutability.is_some()) {
(true, true) => "*mut ",
(true, false) => "*const ",
(false, true) => "&mut ",
(false, false) => "&",
}, this_param).unwrap();
assert!(first_arg);
first_arg = false;
},
syn::FnArg::Typed(arg) => {
if types.skip_arg(&*arg.ty, generics) { continue; }
if !arg.attrs.is_empty() { unimplemented!(); }
let mut is_ref = if let syn::Type::Reference(_) = *arg.ty { true } else { false };
if let syn::Type::Reference(syn::TypeReference { ref elem, .. }) = *arg.ty {
if let syn::Type::Slice(_) = &**elem {
// Slices are mapped to non-ref Vec types, so we want them to be mut
// letting us drain(..) the underlying Vec.
is_ref = false;
}
}
match &*arg.pat {
syn::Pat::Ident(ident) => {
if !ident.attrs.is_empty() || ident.subpat.is_some() {
unimplemented!();
}
write!(w, "{}{}{}: ", if first_arg { "" } else { ", " }, if is_ref || !fn_decl { "" } else { "mut " }, ident.ident).unwrap();
first_arg = false;
},
syn::Pat::Wild(wild) => {
if !wild.attrs.is_empty() { unimplemented!(); }
write!(w, "{}unused_{}: ", if first_arg { "" } else { ", " }, num_unused).unwrap();
num_unused += 1;
},
_ => unimplemented!(),
}
types.write_c_type(w, &*arg.ty, generics, false);
}
}
}
write!(w, ")").unwrap();
match &sig.output {
syn::ReturnType::Type(_, rtype) => {
write!(w, " -> ").unwrap();
if let Some(mut remaining_path) = first_seg_self(&*rtype) {
if let Some(associated_seg) = get_single_remaining_path_seg(&mut remaining_path) {
// We're returning an associated type in a trait impl. Its probably a safe bet
// that its also a trait, so just return the trait type.
let real_type = associated_types.get(associated_seg).unwrap();
types.write_c_type(w, &syn::Type::Path(syn::TypePath { qself: None,
path: syn::PathSegment {
ident: (*real_type).clone(),
arguments: syn::PathArguments::None
}.into()
}), generics, true);
} else {
write!(w, "{}", this_param).unwrap();
}
} else {
if let syn::Type::Reference(r) = &**rtype {
// We can't return a reference, cause we allocate things on the stack.
types.write_c_type(w, &*r.elem, generics, true);
} else {
types.write_c_type(w, &*rtype, generics, true);
}
}
},
_ => {},
}
}
/// Print the main part of a method declaration body, starting with a newline after the function
/// open bracket and converting each function parameter to or from C-mapped types. Ends with "let
/// mut ret = " assuming the next print will be the unmapped Rust function to call followed by the
/// parameters we mapped to/from C here.
pub fn write_method_var_decl_body<W: std::io::Write>(w: &mut W, sig: &syn::Signature, extra_indent: &str, types: &TypeResolver, generics: Option<&GenericTypes>, to_c: bool) {
let mut num_unused = 0;
for inp in sig.inputs.iter() {
match inp {
syn::FnArg::Receiver(_) => {},
syn::FnArg::Typed(arg) => {
if types.skip_arg(&*arg.ty, generics) { continue; }
if !arg.attrs.is_empty() { unimplemented!(); }
macro_rules! write_new_var {
($ident: expr, $ty: expr) => {
if to_c {
if types.write_to_c_conversion_new_var(w, &$ident, &$ty, generics, false) {
write!(w, "\n\t{}", extra_indent).unwrap();
}
} else {
if types.write_from_c_conversion_new_var(w, &$ident, &$ty, generics) {
write!(w, "\n\t{}", extra_indent).unwrap();
}
}
}
}
match &*arg.pat {
syn::Pat::Ident(ident) => {
if !ident.attrs.is_empty() || ident.subpat.is_some() {
unimplemented!();
}
write_new_var!(ident.ident, *arg.ty);
},
syn::Pat::Wild(w) => {
if !w.attrs.is_empty() { unimplemented!(); }
write_new_var!(syn::Ident::new(&format!("unused_{}", num_unused), Span::call_site()), *arg.ty);
num_unused += 1;
},
_ => unimplemented!(),
}
}
}
}
match &sig.output {
syn::ReturnType::Type(_, _) => {
write!(w, "let mut ret = ").unwrap();
},
_ => {},
}
}
/// Prints the parameters in a method call, starting after the open parenthesis and ending with a
/// final return statement returning the method's result. Should be followed by a single closing
/// bracket.
///
/// The return value is expected to be bound to a variable named `ret` which is available after a
/// method-call-ending semicolon.
pub fn write_method_call_params<W: std::io::Write>(w: &mut W, sig: &syn::Signature, associated_types: &HashMap<&syn::Ident, &syn::Ident>, extra_indent: &str, types: &TypeResolver, generics: Option<&GenericTypes>, this_type: &str, to_c: bool) {
let mut first_arg = true;
let mut num_unused = 0;
for inp in sig.inputs.iter() {
match inp {
syn::FnArg::Receiver(recv) => {
if !recv.attrs.is_empty() || recv.reference.is_none() { unimplemented!(); }
if to_c {
write!(w, "self.this_arg").unwrap();
first_arg = false;
}
},
syn::FnArg::Typed(arg) => {
if types.skip_arg(&*arg.ty, generics) {
if !to_c {
if !first_arg {
write!(w, ", ").unwrap();
}
first_arg = false;
types.no_arg_to_rust(w, &*arg.ty, generics);
}
continue;
}
if !arg.attrs.is_empty() { unimplemented!(); }
macro_rules! write_ident {
($ident: expr) => {
if !first_arg {
write!(w, ", ").unwrap();
}
first_arg = false;
if to_c {
types.write_to_c_conversion_inline_prefix(w, &*arg.ty, generics, false);
write!(w, "{}", $ident).unwrap();
types.write_to_c_conversion_inline_suffix(w, &*arg.ty, generics, false);
} else {
types.write_from_c_conversion_prefix(w, &*arg.ty, generics);
write!(w, "{}", $ident).unwrap();
types.write_from_c_conversion_suffix(w, &*arg.ty, generics);
}
}
}
match &*arg.pat {
syn::Pat::Ident(ident) => {
if !ident.attrs.is_empty() || ident.subpat.is_some() {
unimplemented!();
}
write_ident!(ident.ident);
},
syn::Pat::Wild(w) => {
if !w.attrs.is_empty() { unimplemented!(); }
write_ident!(format!("unused_{}", num_unused));
num_unused += 1;
},
_ => unimplemented!(),
}
}
}
}
write!(w, ")").unwrap();
match &sig.output {
syn::ReturnType::Type(_, rtype) => {
write!(w, ";\n\t{}", extra_indent).unwrap();
if to_c && first_seg_self(&*rtype).is_some() {
// Assume rather blindly that we're returning an associated trait from a C fn call to a Rust trait object.
write!(w, "ret").unwrap();
} else if !to_c && first_seg_self(&*rtype).is_some() {
if let Some(mut remaining_path) = first_seg_self(&*rtype) {
if let Some(associated_seg) = get_single_remaining_path_seg(&mut remaining_path) {
let real_type = associated_types.get(associated_seg).unwrap();
if let Some(t) = types.crate_types.traits.get(&types.maybe_resolve_ident(&real_type).unwrap()) {
// We're returning an associated trait from a Rust fn call to a C trait
// object.
writeln!(w, "let mut rust_obj = {} {{ inner: Box::into_raw(Box::new(ret)), is_owned: true }};", this_type).unwrap();
writeln!(w, "\t{}let mut ret = {}_as_{}(&rust_obj);", extra_indent, this_type, t.ident).unwrap();
writeln!(w, "\t{}// We want to free rust_obj when ret gets drop()'d, not rust_obj, so wipe rust_obj's pointer and set ret's free() fn", extra_indent).unwrap();
writeln!(w, "\t{}rust_obj.inner = std::ptr::null_mut();", extra_indent).unwrap();
writeln!(w, "\t{}ret.free = Some({}_free_void);", extra_indent, this_type).unwrap();
writeln!(w, "\t{}ret", extra_indent).unwrap();
return;
}
}
}
write!(w, "{} {{ inner: Box::into_raw(Box::new(ret)), is_owned: true }}", this_type).unwrap();
} else if to_c {
let new_var = types.write_from_c_conversion_new_var(w, &syn::Ident::new("ret", Span::call_site()), rtype, generics);
if new_var {
write!(w, "\n\t{}", extra_indent).unwrap();
}
types.write_from_c_conversion_prefix(w, &*rtype, generics);
write!(w, "ret").unwrap();
types.write_from_c_conversion_suffix(w, &*rtype, generics);
} else {
let ret_returned = if let syn::Type::Reference(_) = &**rtype { true } else { false };
let new_var = types.write_to_c_conversion_new_var(w, &syn::Ident::new("ret", Span::call_site()), &rtype, generics, true);
if new_var {
write!(w, "\n\t{}", extra_indent).unwrap();
}
types.write_to_c_conversion_inline_prefix(w, &rtype, generics, true);
write!(w, "{}ret", if ret_returned && !new_var { "*" } else { "" }).unwrap();
types.write_to_c_conversion_inline_suffix(w, &rtype, generics, true);
}
}
_ => {},
}
}
/// Prints concrete generic parameters for a struct/trait/function, including the less-than and
/// greater-than symbols, if any generic parameters are defined.
pub fn maybe_write_generics<W: std::io::Write>(w: &mut W, generics: &syn::Generics, types: &TypeResolver, concrete_lifetimes: bool) {
let mut gen_types = GenericTypes::new();
assert!(gen_types.learn_generics(generics, types));
if !generics.params.is_empty() {
write!(w, "<").unwrap();
for (idx, generic) in generics.params.iter().enumerate() {
match generic {
syn::GenericParam::Type(type_param) => {
let mut printed_param = false;
for bound in type_param.bounds.iter() {
if let syn::TypeParamBound::Trait(trait_bound) = bound {
assert_simple_bound(&trait_bound);
write!(w, "{}{}", if idx != 0 { ", " } else { "" }, gen_types.maybe_resolve_ident(&type_param.ident).unwrap()).unwrap();
if printed_param {
unimplemented!("Can't print generic params that have multiple non-lifetime bounds");
}
printed_param = true;
}
}
},
syn::GenericParam::Lifetime(lt) => {
if concrete_lifetimes {
write!(w, "'static").unwrap();
} else {
write!(w, "{}'{}", if idx != 0 { ", " } else { "" }, lt.lifetime.ident).unwrap();
}
},
_ => unimplemented!(),
}
}
write!(w, ">").unwrap();
}
}

1283
c-bindings-gen/src/main.rs Normal file

File diff suppressed because it is too large Load diff

2021
c-bindings-gen/src/types.rs Normal file

File diff suppressed because it is too large Load diff