Merge pull request #110 from Nadrieril/parser-proc-macro

Rewrite parser macros as proc_macros

3 files changed, 342 insertions, 0 deletions

diff --git a/dhall_proc_macros/src/lib.rs b/dhall_proc_macros/src/lib.rs
index 5304429..37e8f9f 100644
--- a/dhall_proc_macros/src/lib.rs
+++ b/dhall_proc_macros/src/lib.rs
@@ -1,3 +1,4 @@
+#![feature(drain_filter)]
 //! This crate contains the code-generation primitives for the [dhall-rust][dhall-rust] crate.
 //! This is highly unstable and breaks regularly; use at your own risk.
 //!
@@ -6,6 +7,8 @@
 extern crate proc_macro;
 
 mod derive;
+mod make_parser;
+mod parse_children;
 
 use proc_macro::TokenStream;
 
@@ -13,3 +16,19 @@ use proc_macro::TokenStream;
 pub fn derive_static_type(input: TokenStream) -> TokenStream {
     derive::derive_static_type(input)
 }
+
+#[proc_macro_attribute]
+pub fn make_parser(attrs: TokenStream, input: TokenStream) -> TokenStream {
+    TokenStream::from(match make_parser::make_parser(attrs, input) {
+        Ok(tokens) => tokens,
+        Err(err) => err.to_compile_error(),
+    })
+}
+
+#[proc_macro]
+pub fn parse_children(input: TokenStream) -> TokenStream {
+    TokenStream::from(match parse_children::parse_children(input) {
+        Ok(tokens) => tokens,
+        Err(err) => err.to_compile_error(),
+    })
+}
diff --git a/dhall_proc_macros/src/make_parser.rs b/dhall_proc_macros/src/make_parser.rs
new file mode 100644
index 0000000..268a639
--- /dev/null
+++ b/dhall_proc_macros/src/make_parser.rs
@@ -0,0 +1,111 @@
+use quote::quote;
+use syn::parse::{ParseStream, Result};
+use syn::spanned::Spanned;
+use syn::{
+    parse_quote, Error, Expr, FnArg, Ident, ImplItem, ImplItemMethod, ItemImpl,
+    Pat, Token,
+};
+
+fn apply_special_attrs(function: &mut ImplItemMethod) -> Result<()> {
+    *function = parse_quote!(
+        #[allow(non_snake_case, dead_code)]
+        #function
+    );
+
+    let recognized_attrs: Vec<_> = function
+        .attrs
+        .drain_filter(|attr| attr.path.is_ident("prec_climb"))
+        .collect();
+
+    let name = function.sig.ident.clone();
+
+    if recognized_attrs.is_empty() {
+        // do nothing
+    } else if recognized_attrs.len() > 1 {
+        return Err(Error::new(
+            recognized_attrs[1].span(),
+            "expected a single prec_climb attribute",
+        ));
+    } else {
+        let attr = recognized_attrs.into_iter().next().unwrap();
+        let (child_rule, climber) =
+            attr.parse_args_with(|input: ParseStream| {
+                let child_rule: Ident = input.parse()?;
+                let _: Token![,] = input.parse()?;
+                let climber: Expr = input.parse()?;
+                Ok((child_rule, climber))
+            })?;
+
+        // Get the name of the first (`input`) function argument
+        let first_arg = function.sig.inputs.first().ok_or_else(|| {
+            Error::new(
+                function.sig.inputs.span(),
+                "a prec_climb function needs 4 arguments",
+            )
+        })?;
+        let first_arg = match &first_arg {
+            FnArg::Receiver(_) => return Err(Error::new(
+                first_arg.span(),
+                "a prec_climb function should not have a `self` argument",
+            )),
+            FnArg::Typed(first_arg) =>   match &*first_arg.pat{
+                Pat::Ident(ident) => &ident.ident,
+                _ => return Err(Error::new(
+                    first_arg.span(),
+                    "this argument should be a plain identifier instead of a pattern",
+                )),
+            }
+        };
+
+        function.block = parse_quote!({
+            #function
+
+            #climber.climb(
+                #first_arg.pair.clone().into_inner(),
+                |p| Self::#child_rule(#first_arg.with_pair(p)),
+                |l, op, r| {
+                    #name(#first_arg.clone(), l?, op, r?)
+                },
+            )
+        });
+        // Remove the 3 last arguments to keep only the `input` one
+        function.sig.inputs.pop();
+        function.sig.inputs.pop();
+        function.sig.inputs.pop();
+        // Check that an argument remains
+        function.sig.inputs.first().ok_or_else(|| {
+            Error::new(
+                function.sig.inputs.span(),
+                "a prec_climb function needs 4 arguments",
+            )
+        })?;
+    }
+
+    Ok(())
+}
+
+pub fn make_parser(
+    attrs: proc_macro::TokenStream,
+    input: proc_macro::TokenStream,
+) -> Result<proc_macro2::TokenStream> {
+    let rule_enum: Ident = syn::parse(attrs)?;
+
+    let mut imp: ItemImpl = syn::parse(input)?;
+    imp.items
+        .iter_mut()
+        .map(|item| match item {
+            ImplItem::Method(m) => apply_special_attrs(m),
+            _ => Ok(()),
+        })
+        .collect::<Result<()>>()?;
+
+    let ty = &imp.self_ty;
+    let (impl_generics, _, where_clause) = imp.generics.split_for_impl();
+    Ok(quote!(
+        impl #impl_generics PestConsumer for #ty #where_clause {
+            type RuleEnum = #rule_enum;
+        }
+
+        #imp
+    ))
+}
diff --git a/dhall_proc_macros/src/parse_children.rs b/dhall_proc_macros/src/parse_children.rs
new file mode 100644
index 0000000..c78adbc
--- /dev/null
+++ b/dhall_proc_macros/src/parse_children.rs
@@ -0,0 +1,212 @@
+use proc_macro2::{Span, TokenStream};
+use quote::quote;
+use syn::parse::{Parse, ParseStream, Result};
+use syn::punctuated::Punctuated;
+use syn::spanned::Spanned;
+use syn::{bracketed, parenthesized, token, Error, Expr, Ident, Pat, Token};
+
+#[derive(Debug, Clone)]
+struct ChildrenBranch {
+    pattern_span: Span,
+    pattern: Punctuated<ChildrenBranchPatternItem, Token![,]>,
+    body: Expr,
+}
+
+#[derive(Debug, Clone)]
+enum ChildrenBranchPatternItem {
+    Single { rule_name: Ident, binder: Pat },
+    Multiple { rule_name: Ident, binder: Ident },
+}
+
+#[derive(Debug, Clone)]
+struct ParseChildrenInput {
+    input_expr: Expr,
+    branches: Punctuated<ChildrenBranch, Token![,]>,
+}
+
+impl Parse for ChildrenBranch {
+    fn parse(input: ParseStream) -> Result<Self> {
+        let contents;
+        let _: token::Bracket = bracketed!(contents in input);
+        let pattern_unparsed: TokenStream = contents.fork().parse()?;
+        let pattern_span = pattern_unparsed.span();
+        let pattern = Punctuated::parse_terminated(&contents)?;
+        let _: Token![=>] = input.parse()?;
+        let body = input.parse()?;
+
+        Ok(ChildrenBranch {
+            pattern_span,
+            pattern,
+            body,
+        })
+    }
+}
+
+impl Parse for ChildrenBranchPatternItem {
+    fn parse(input: ParseStream) -> Result<Self> {
+        let contents;
+        let rule_name = input.parse()?;
+        parenthesized!(contents in input);
+        if input.peek(Token![..]) {
+            let binder = contents.parse()?;
+            let _: Token![..] = input.parse()?;
+            Ok(ChildrenBranchPatternItem::Multiple { rule_name, binder })
+        } else if input.is_empty() || input.peek(Token![,]) {
+            let binder = contents.parse()?;
+            Ok(ChildrenBranchPatternItem::Single { rule_name, binder })
+        } else {
+            Err(input.error("expected `..` or nothing"))
+        }
+    }
+}
+
+impl Parse for ParseChildrenInput {
+    fn parse(input: ParseStream) -> Result<Self> {
+        let input_expr = input.parse()?;
+        let _: Token![;] = input.parse()?;
+        let branches = Punctuated::parse_terminated(input)?;
+
+        Ok(ParseChildrenInput {
+            input_expr,
+            branches,
+        })
+    }
+}
+
+fn make_parser_branch(
+    branch: &ChildrenBranch,
+    i_inputs: &Ident,
+) -> Result<TokenStream> {
+    use ChildrenBranchPatternItem::{Multiple, Single};
+
+    let body = &branch.body;
+
+    // Convert the input pattern into a pattern-match on the Rules of the children. This uses
+    // slice_patterns.
+    // A single pattern just checks that the rule matches; a variable-length pattern binds the
+    // subslice and checks, in the if-guard, that its elements all match the chosen Rule.
+    let i_variable_pattern =
+        Ident::new("___variable_pattern", Span::call_site());
+    let match_pat = branch.pattern.iter().map(|item| match item {
+        Single { rule_name, .. } => {
+            quote!(<<Self as PestConsumer>::RuleEnum>::#rule_name)
+        }
+        Multiple { .. } => quote!(#i_variable_pattern..),
+    });
+    let match_filter = branch.pattern.iter().map(|item| match item {
+        Single { .. } => quote!(),
+        Multiple { rule_name, .. } => quote!(
+            {
+                // We can't use .all() directly in the pattern guard; see
+                // https://github.com/rust-lang/rust/issues/59803.
+                let all_match = |slice: &[_]| {
+                    slice.iter().all(|r|
+                        r == &<<Self as PestConsumer>::RuleEnum>::#rule_name
+                    )
+                };
+                all_match(#i_variable_pattern)
+            } &&
+        ),
+    });
+
+    // Once we have found a branch that matches, we need to parse the children.
+    let mut singles_before_multiple = Vec::new();
+    let mut multiple = None;
+    let mut singles_after_multiple = Vec::new();
+    for item in &branch.pattern {
+        match item {
+            Single {
+                rule_name, binder, ..
+            } => {
+                if multiple.is_none() {
+                    singles_before_multiple.push((rule_name, binder))
+                } else {
+                    singles_after_multiple.push((rule_name, binder))
+                }
+            }
+            Multiple {
+                rule_name, binder, ..
+            } => {
+                if multiple.is_none() {
+                    multiple = Some((rule_name, binder))
+                } else {
+                    return Err(Error::new(
+                        branch.pattern_span.clone(),
+                        "multiple variable-length patterns are not allowed",
+                    ));
+                }
+            }
+        }
+    }
+    let mut parses = Vec::new();
+    for (rule_name, binder) in singles_before_multiple.into_iter() {
+        parses.push(quote!(
+            let #binder = Self::#rule_name(
+                #i_inputs.next().unwrap()
+            )?;
+        ))
+    }
+    // Note the `rev()`: we are taking inputs from the end of the iterator in reverse order, so that
+    // only the unmatched inputs are left for the variable-length pattern, if any.
+    for (rule_name, binder) in singles_after_multiple.into_iter().rev() {
+        parses.push(quote!(
+            let #binder = Self::#rule_name(
+                #i_inputs.next_back().unwrap()
+            )?;
+        ))
+    }
+    if let Some((rule_name, binder)) = multiple {
+        parses.push(quote!(
+            let #binder = #i_inputs
+                .map(|i| Self::#rule_name(i))
+                .collect::<Result<Vec<_>, _>>()?
+                .into_iter();
+        ))
+    }
+
+    Ok(quote!(
+        [#(#match_pat),*] if #(#match_filter)* true => {
+            #(#parses)*
+            #body
+        }
+    ))
+}
+
+pub fn parse_children(
+    input: proc_macro::TokenStream,
+) -> Result<proc_macro2::TokenStream> {
+    let input: ParseChildrenInput = syn::parse(input)?;
+
+    let i_children_rules = Ident::new("___children_rules", Span::call_site());
+    let i_inputs = Ident::new("___inputs", Span::call_site());
+
+    let input_expr = &input.input_expr;
+    let branches = input
+        .branches
+        .iter()
+        .map(|br| make_parser_branch(br, &i_inputs))
+        .collect::<Result<Vec<_>>>()?;
+
+    Ok(quote!({
+        let #i_children_rules: Vec<_> = #input_expr.pair
+            .clone()
+            .into_inner()
+            .map(|p| p.as_rule())
+            .collect();
+
+        #[allow(unused_mut)]
+        let mut #i_inputs = #input_expr
+            .pair
+            .clone()
+            .into_inner()
+            .map(|p| #input_expr.with_pair(p));
+
+        #[allow(unreachable_code)]
+        match #i_children_rules.as_slice() {
+            #(#branches,)*
+            [..] => return Err(#input_expr.error(
+                format!("Unexpected children: {:?}", #i_children_rules)
+            )),
+        }
+    }))
+}