From f895205d41ced8ddc7b9f72fdf9e474292aa2cb8 Mon Sep 17 00:00:00 2001 From: David Tolnay Date: Fri, 19 Apr 2024 18:46:16 -0700 Subject: [PATCH] Add classify::expr_is_complete --- compiler/rustc_ast/src/util/classify.rs | 99 +++++++++++++++---------- compiler/rustc_parse/src/parser/expr.rs | 62 +--------------- 2 files changed, 61 insertions(+), 100 deletions(-) diff --git a/compiler/rustc_ast/src/util/classify.rs b/compiler/rustc_ast/src/util/classify.rs index 86383af1f7c83..3e296cb1c38ab 100644 --- a/compiler/rustc_ast/src/util/classify.rs +++ b/compiler/rustc_ast/src/util/classify.rs @@ -1,12 +1,60 @@ //! Routines the parser and pretty-printer use to classify AST nodes. +use crate::ast::ExprKind::*; use crate::{ast, token::Delimiter}; +/// This classification determines whether various syntactic positions break out +/// of parsing the current expression (true) or continue parsing more of the +/// same expression (false). +/// +/// For example, it's relevant in the parsing of match arms: +/// +/// ```ignore (illustrative) +/// match ... { +/// // Is this calling $e as a function, or is it the start of a new arm +/// // with a tuple pattern? +/// _ => $e ( +/// ^ ) +/// +/// // Is this an Index operation, or new arm with a slice pattern? +/// _ => $e [ +/// ^ ] +/// +/// // Is this a binary operator, or leading vert in a new arm? Same for +/// // other punctuation which can either be a binary operator in +/// // expression or unary operator in pattern, such as `&` and `-`. +/// _ => $e | +/// ^ +/// } +/// ``` +/// +/// If $e is something like `{}` or `if … {}`, then terminate the current +/// arm and parse a new arm. +/// +/// If $e is something like `path::to` or `(…)`, continue parsing the same +/// arm. +/// +/// *Almost* the same classification is used as an early bail-out for parsing +/// statements. See `expr_requires_semi_to_be_stmt`. +pub fn expr_is_complete(e: &ast::Expr) -> bool { + matches!( + e.kind, + If(..) + | Match(..) + | Block(..) + | While(..) + | Loop(..) + | ForLoop { .. } + | TryBlock(..) + | ConstBlock(..) + ) +} + /// Does this expression require a semicolon to be treated as a statement? /// /// The negation of this: "can this expression be used as a statement without a -/// semicolon" -- is used as an early bail-out in the parser so that, for -/// instance, +/// semicolon" -- is used as an early bail-out when parsing statements so that, +/// for instance, /// /// ```ignore (illustrative) /// if true {...} else {...} @@ -15,56 +63,25 @@ use crate::{ast, token::Delimiter}; /// /// isn't parsed as `(if true {...} else {...} | x) | 5`. /// -/// Nearly the same early bail-out also occurs in the right-hand side of match -/// arms: -/// -/// ```ignore (illustrative) -/// match i { -/// 0 => if true {...} else {...} -/// | x => {} -/// } -/// ``` -/// -/// Here the `|` is a leading vert in a second match arm. It is not a binary -/// operator with the If as its left operand. If the first arm were some other -/// expression for which `expr_requires_semi_to_be_stmt` returns true, then the -/// `|` on the next line would be a binary operator (leading to a parse error). +/// Surprising special case: even though braced macro calls like `m! {}` +/// normally do not introduce a boundary when found at the head of a match arm, +/// they do terminate the parsing of a statement. /// -/// The statement case and the match-arm case are "nearly" the same early -/// bail-out because of 1 edge case. Macro calls with brace delimiter terminate -/// a statement without a semicolon, but do not terminate a match-arm without -/// comma. +/// match ... { +/// _ => m! {} (), // macro that expands to a function, which is then called +/// } /// -/// ```ignore (illustrative) -/// m! {} - 1; // two statements: a macro call followed by -1 literal +/// let _ = { m! {} () }; // macro call followed by unit /// -/// match () { -/// _ => m! {} - 1, // binary subtraction operator -/// } -/// ``` pub fn expr_requires_semi_to_be_stmt(e: &ast::Expr) -> bool { - use ast::ExprKind::*; - match &e.kind { - If(..) - | Match(..) - | Block(..) - | While(..) - | Loop(..) - | ForLoop { .. } - | TryBlock(..) - | ConstBlock(..) => false, - MacCall(mac_call) => mac_call.args.delim != Delimiter::Brace, - - _ => true, + _ => !expr_is_complete(e), } } /// If an expression ends with `}`, returns the innermost expression ending in the `}` pub fn expr_trailing_brace(mut expr: &ast::Expr) -> Option<&ast::Expr> { - use ast::ExprKind::*; - loop { match &expr.kind { AddrOf(_, _, e) diff --git a/compiler/rustc_parse/src/parser/expr.rs b/compiler/rustc_parse/src/parser/expr.rs index 52e3e33691a13..441aa5b0806da 100644 --- a/compiler/rustc_parse/src/parser/expr.rs +++ b/compiler/rustc_parse/src/parser/expr.rs @@ -496,51 +496,8 @@ impl<'a> Parser<'a> { } /// Checks if this expression is a successfully parsed statement. - /// - /// This determines whether to continue parsing more of an expression in a - /// match arm (false) vs continue to the next arm (true). - /// - /// ```ignore (illustrative) - /// match ... { - /// // Is this calling $e as a function, or is it the start of a new arm - /// // with a tuple pattern? - /// _ => $e ( - /// ^ ) - /// - /// // Is this an Index operation, or new arm with a slice pattern? - /// _ => $e [ - /// ^ ] - /// - /// // Is this a binary operator, or leading vert in a new arm? Same for - /// // other punctuation which can either be a binary operator in - /// // expression or unary operator in pattern, such as `&` and `-`. - /// _ => $e | - /// ^ - /// } - /// ``` - /// - /// If $e is something like `path::to` or `(…)`, continue parsing the same - /// arm. - /// - /// If $e is something like `{}` or `if … {}`, then terminate the current - /// arm and parse a new arm. fn expr_is_complete(&self, e: &Expr) -> bool { - self.restrictions.contains(Restrictions::STMT_EXPR) - && match e.kind { - // Surprising special case: even though braced macro calls like - // `m! {}` normally introduce a statement boundary when found at - // the head of a statement, in match arms they do not terminate - // the arm. - // - // let _ = { m! {} () }; // macro call followed by unit - // - // match ... { - // _ => m! {} (), // macro that expands to a function, which is then called - // } - // - ExprKind::MacCall(_) => false, - _ => !classify::expr_requires_semi_to_be_stmt(e), - } + self.restrictions.contains(Restrictions::STMT_EXPR) && classify::expr_is_complete(e) } /// Parses `x..y`, `x..=y`, and `x..`/`x..=`. @@ -3203,21 +3160,8 @@ impl<'a> Parser<'a> { err })?; - let require_comma = match expr.kind { - // Special case: braced macro calls require comma in a match - // arm, even though they do not require semicolon in a - // statement. - // - // m! {} // okay without semicolon - // - // match ... { - // _ => m! {}, // requires comma - // _ => ... - // } - // - ExprKind::MacCall(_) => true, - _ => classify::expr_requires_semi_to_be_stmt(&expr), - } && this.token != token::CloseDelim(Delimiter::Brace); + let require_comma = !classify::expr_is_complete(&expr) + && this.token != token::CloseDelim(Delimiter::Brace); if !require_comma { arm_body = Some(expr);