use itertools::Itertools; use pest::iterators::Pair; use pest::Parser; use std::borrow::Cow; use std::rc::Rc; use dhall_generated_parser::{DhallParser, Rule}; use crate::map::DupTreeMap; use crate::ExprF::*; use crate::*; // This file consumes the parse tree generated by pest and turns it into // our own AST. All those custom macros should eventually moved into // their own crate because they are quite general and useful. For now they // are here and hopefully you can figure out how they work. type ParsedExpr = Expr; type ParsedSubExpr = SubExpr; type ParsedText = InterpolatedText>; type ParsedTextContents = InterpolatedTextContents>; pub type ParseError = pest::error::Error; pub type ParseResult = Result; fn unspanned(x: ParsedExpr) -> ParsedSubExpr { SubExpr::from_expr_no_note(x) } #[derive(Debug, Clone)] pub struct Span { input: Rc, /// # Safety /// /// Must be a valid character boundary index into `input`. start: usize, /// # Safety /// /// Must be a valid character boundary index into `input`. end: usize, } impl Span { fn make(input: Rc, sp: pest::Span) -> Self { Span { input, start: sp.start(), end: sp.end(), } } } fn spanned(span: Span, x: ParsedExpr) -> ParsedSubExpr { SubExpr::new(x, span) } #[derive(Debug)] enum Either { Left(A), Right(B), } impl crate::Builtin { pub fn parse(s: &str) -> Option { use crate::Builtin::*; match s { "Bool" => Some(Bool), "Natural" => Some(Natural), "Integer" => Some(Integer), "Double" => Some(Double), "Text" => Some(Text), "List" => Some(List), "Optional" => Some(Optional), "None" => Some(OptionalNone), "Natural/build" => Some(NaturalBuild), "Natural/fold" => Some(NaturalFold), "Natural/isZero" => Some(NaturalIsZero), "Natural/even" => Some(NaturalEven), "Natural/odd" => Some(NaturalOdd), "Natural/toInteger" => Some(NaturalToInteger), "Natural/show" => Some(NaturalShow), "Integer/toDouble" => Some(IntegerToDouble), "Integer/show" => Some(IntegerShow), "Double/show" => Some(DoubleShow), "List/build" => Some(ListBuild), "List/fold" => Some(ListFold), "List/length" => Some(ListLength), "List/head" => Some(ListHead), "List/last" => Some(ListLast), "List/indexed" => Some(ListIndexed), "List/reverse" => Some(ListReverse), "Optional/fold" => Some(OptionalFold), "Optional/build" => Some(OptionalBuild), "Text/show" => Some(TextShow), _ => None, } } } pub fn custom_parse_error(pair: &Pair, msg: String) -> ParseError { let msg = format!("{} while matching on:\n{}", msg, debug_pair(pair.clone())); let e = pest::error::ErrorVariant::CustomError { message: msg }; pest::error::Error::new_from_span(e, pair.as_span()) } fn debug_pair(pair: Pair) -> String { use std::fmt::Write; let mut s = String::new(); fn aux(s: &mut String, indent: usize, prefix: String, pair: Pair) { let indent_str = "| ".repeat(indent); let rule = pair.as_rule(); let contents = pair.as_str(); let mut inner = pair.into_inner(); let mut first = true; while let Some(p) = inner.next() { if first { first = false; let last = inner.peek().is_none(); if last && p.as_str() == contents { let prefix = format!("{}{:?} > ", prefix, rule); aux(s, indent, prefix, p); continue; } else { writeln!( s, r#"{}{}{:?}: "{}""#, indent_str, prefix, rule, contents ) .unwrap(); } } aux(s, indent + 1, "".into(), p); } if first { writeln!( s, r#"{}{}{:?}: "{}""#, indent_str, prefix, rule, contents ) .unwrap(); } } aux(&mut s, 0, "".into(), pair); s } macro_rules! make_parser { (@pattern, rule, $name:ident) => (Rule::$name); (@pattern, token_rule, $name:ident) => (Rule::$name); (@pattern, rule_group, $name:ident) => (_); (@filter, rule) => (true); (@filter, token_rule) => (true); (@filter, rule_group) => (false); (@body, ($($things:tt)*), rule!( $name:ident<$o:ty>; $($args:tt)* ) ) => ( make_parser!(@body, ($($things)*), rule!( $name<$o> as $name; $($args)* ) ) ); (@body, ($_input:expr, $pair:expr, $_children:expr), rule!( $name:ident<$o:ty> as $group:ident; captured_str!($x:pat) => $body:expr ) ) => ({ let $x = $pair.as_str(); let res: $o = $body; Ok(ParsedValue::$group(res)) }); (@body, ($_input:expr, $_pair:expr, $children:expr), rule!( $name:ident<$o:ty> as $group:ident; children!( $( [$($args:tt)*] => $body:expr ),* $(,)* ) ) ) => ({ #[allow(unused_imports)] use ParsedValue::*; #[allow(unreachable_code)] let res: $o = improved_slice_patterns::match_vec!($children; $( [$($args)*] => $body, )* [x..] => Err( format!("Unexpected children: {:?}", x.collect::>()) )?, ).map_err(|_| -> String { unreachable!() })?; Ok(ParsedValue::$group(res)) }); (@body, ($input:expr, $pair:expr, $children:expr), rule!( $name:ident<$o:ty> as $group:ident; $span:ident; $($args:tt)* ) ) => ({ let $span = Span::make($input, $pair.as_span()); make_parser!(@body, ($input, $pair, $children), rule!( $name<$o> as $group; $($args)* ) ) }); (@body, ($($things:tt)*), token_rule!($name:ident<$o:ty>) ) => ({ Ok(ParsedValue::$name(())) }); (@body, ($($things:tt)*), rule_group!( $name:ident<$o:ty> )) => ( unreachable!() ); ($( $submac:ident!( $name:ident<$o:ty> $($args:tt)* ); )*) => ( #[allow(non_camel_case_types, dead_code, clippy::large_enum_variant)] #[derive(Debug)] enum ParsedValue<'a> { $( $name($o), )* } fn parse_any<'a>( input: Rc, pair: Pair<'a, Rule>, children: Vec>, ) -> Result, String> { match pair.as_rule() { $( make_parser!(@pattern, $submac, $name) if make_parser!(@filter, $submac) => make_parser!(@body, (input, pair, children), $submac!( $name<$o> $($args)* )) , )* r => Err(format!("Unexpected {:?}", r)), } } ); } // Non-recursive implementation to avoid stack overflows fn do_parse<'a>( input: Rc, initial_pair: Pair<'a, Rule>, ) -> ParseResult> { enum StackFrame<'a> { Unprocessed(Pair<'a, Rule>), Processed(Pair<'a, Rule>, usize), } use StackFrame::*; let mut pairs_stack: Vec = vec![Unprocessed(initial_pair.clone())]; let mut values_stack: Vec = vec![]; while let Some(p) = pairs_stack.pop() { match p { Unprocessed(mut pair) => loop { let mut pairs: Vec<_> = pair.clone().into_inner().collect(); let n_children = pairs.len(); if n_children == 1 && can_be_shortcutted(pair.as_rule()) { pair = pairs.pop().unwrap(); continue; } else { pairs_stack.push(Processed(pair, n_children)); pairs_stack .extend(pairs.into_iter().map(StackFrame::Unprocessed)); break; } }, Processed(pair, n) => { let mut children: Vec<_> = values_stack.split_off(values_stack.len() - n); children.reverse(); let val = match parse_any(input.clone(), pair.clone(), children) { Ok(v) => v, Err(msg) => Err(custom_parse_error(&pair, msg))?, }; values_stack.push(val); } } } Ok(values_stack.pop().unwrap()) } // List of rules that can be shortcutted if they have a single child fn can_be_shortcutted(rule: Rule) -> bool { use Rule::*; match rule { expression | import_alt_expression | or_expression | plus_expression | text_append_expression | list_append_expression | and_expression | combine_expression | prefer_expression | combine_types_expression | times_expression | equal_expression | not_equal_expression | application_expression | first_application_expression | selector_expression | annotated_expression => true, _ => false, } } // Trim the shared indent off of a vec of lines, as defined by the Dhall semantics of multiline // literals. fn trim_indent(lines: &mut Vec) { let is_indent = |c: char| c == ' ' || c == '\t'; // There is at least one line so this is safe let last_line_head = lines.last().unwrap().head(); let indent_chars = last_line_head .char_indices() .take_while(|(_, c)| is_indent(*c)); let mut min_indent_idx = match indent_chars.last() { Some((i, _)) => i, // If there is no indent char, then no indent needs to be stripped None => return, }; for line in lines.iter() { // Ignore empty lines if line.is_empty() { continue; } // Take chars from line while they match the current minimum indent. let indent_chars = last_line_head[0..=min_indent_idx] .char_indices() .zip(line.head().chars()) .take_while(|((_, c1), c2)| c1 == c2); match indent_chars.last() { Some(((i, _), _)) => min_indent_idx = i, // If there is no indent char, then no indent needs to be stripped None => return, }; } // Remove the shared indent from non-empty lines for line in lines.iter_mut() { if !line.is_empty() { line.head_mut().replace_range(0..=min_indent_idx, ""); } } } make_parser! { token_rule!(EOI<()>); rule!(simple_label