Auto merge of rust-lang#119427 - dtolnay:maccall, r=compiler-errors

Fix, document, and test parser and pretty-printer edge cases related to braced macro calls

_Review note: this is a deceptively small PR because it comes with 145 lines of docs and 196 lines of tests, and only 25 lines of compiler code changed. However, I recommend reviewing it 1 commit at a time because much of the effect of the code changes is non-local i.e. affecting code that is not visible in the final state of the PR. I have paid attention that reviewing the PR one commit at a time is as easy as I can make it. All of the code you need to know about is touched in those commits, even if some of those changes disappear by the end of the stack._

This is a follow-up to rust-lang#119105. One case that is not relevant to `-Zunpretty=expanded`, but which came up as I'm porting rust-lang#119105 and rust-lang#118726 into `syn`'s printer and `prettyplease`'s printer where it **is** relevant, and is also relevant to rustc's `stringify!`, is statement boundaries in the vicinity of braced macro calls.

Rustc's AST pretty-printer produces invalid syntax for statements that begin with a braced macro call:

```rust
macro_rules! stringify_item {
    ($i:item) => {
        stringify!($i)
    };
}

macro_rules! repro {
    ($e:expr) => {
        stringify_item!(fn main() { $e + 1; })
    };
}

fn main() {
    println!("{}", repro!(m! {}));
}
```

**Before this PR:** output is not valid Rust syntax.

```console
fn main() { m! {} + 1; }
```

```console
error: leading `+` is not supported
 --> <anon>:1:19
  |
1 | fn main() { m! {} + 1; }
  |                   ^ unexpected `+`
  |
help: try removing the `+`
  |
1 - fn main() { m! {} + 1; }
1 + fn main() { m! {}  1; }
  |
```

**After this PR:** valid syntax.

```console
fn main() { (m! {}) + 1; }
```

Author: bors

Diff for: compiler/rustc_ast/src/util/classify.rs

@@ -1,34 +1,88 @@
-//! Routines the parser uses to classify AST nodes
-
-// Predicates on exprs and stmts that the pretty-printer and parser use
+//! Routines the parser and pretty-printer use to classify AST nodes.
 
+use crate::ast::ExprKind::*;
 use crate::{ast, token::Delimiter};
 
-/// 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,
-///     if true {...} else {...}
-///      |x| 5
-/// isn't parsed as (if true {...} else {...} | x) | 5
-pub fn expr_requires_semi_to_be_stmt(e: &ast::Expr) -> bool {
-    !matches!(
+/// 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,
-        ast::ExprKind::If(..)
-            | ast::ExprKind::Match(..)
-            | ast::ExprKind::Block(..)
-            | ast::ExprKind::While(..)
-            | ast::ExprKind::Loop(..)
-            | ast::ExprKind::ForLoop { .. }
-            | ast::ExprKind::TryBlock(..)
-            | ast::ExprKind::ConstBlock(..)
+        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 when parsing statements so that,
+/// for instance,
+///
+/// ```ignore (illustrative)
+/// if true {...} else {...}
+/// |x| 5
+/// ```
+///
+/// isn't parsed as `(if true {...} else {...} | x) | 5`.
+///
+/// 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.
+///
+/// ```ignore (illustrative)
+/// match ... {
+///     _ => m! {} (),  // macro that expands to a function, which is then called
+/// }
+///
+/// let _ = { m! {} () };  // macro call followed by unit
+/// ```
+pub fn expr_requires_semi_to_be_stmt(e: &ast::Expr) -> bool {
+    match &e.kind {
+        MacCall(mac_call) => mac_call.args.delim != Delimiter::Brace,
+        _ => !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 for: compiler/rustc_ast_pretty/src/pprust/state/expr.rs

@@ -780,7 +780,7 @@ impl<'a> State<'a> {
                 }
                 _ => {
                     self.end(); // Close the ibox for the pattern.
-                    self.print_expr(body, FixupContext::new_stmt());
+                    self.print_expr(body, FixupContext::new_match_arm());
                     self.word(",");
                 }
             }

Diff for: compiler/rustc_ast_pretty/src/pprust/state/fixup.rs

@@ -49,6 +49,38 @@ pub(crate) struct FixupContext {
     ///     No parentheses required.
     leftmost_subexpression_in_stmt: bool,
 
+    /// Print expression such that it can be parsed as a match arm.
+    ///
+    /// This is almost equivalent to `stmt`, but the grammar diverges a tiny bit
+    /// between statements and match arms when it comes to braced macro calls.
+    /// Macro calls with brace delimiter terminate a statement without a
+    /// semicolon, but do not terminate a match-arm without comma.
+    ///
+    /// ```ignore (illustrative)
+    /// m! {} - 1;  // two statements: a macro call followed by -1 literal
+    ///
+    /// match () {
+    ///     _ => m! {} - 1,  // binary subtraction operator
+    /// }
+    /// ```
+    match_arm: bool,
+
+    /// This is almost equivalent to `leftmost_subexpression_in_stmt`, other
+    /// than for braced macro calls.
+    ///
+    /// If we have `m! {} - 1` as an expression, the leftmost subexpression
+    /// `m! {}` will need to be parenthesized in the statement case but not the
+    /// match-arm case.
+    ///
+    /// ```ignore (illustrative)
+    /// (m! {}) - 1;  // subexpression needs parens
+    ///
+    /// match () {
+    ///     _ => m! {} - 1,  // no parens
+    /// }
+    /// ```
+    leftmost_subexpression_in_match_arm: bool,
+
     /// This is the difference between:
     ///
     /// ```ignore (illustrative)
@@ -68,6 +100,8 @@ impl Default for FixupContext {
         FixupContext {
             stmt: false,
             leftmost_subexpression_in_stmt: false,
+            match_arm: false,
+            leftmost_subexpression_in_match_arm: false,
             parenthesize_exterior_struct_lit: false,
         }
     }
@@ -76,13 +110,16 @@ impl Default for FixupContext {
 impl FixupContext {
     /// Create the initial fixup for printing an expression in statement
     /// position.
-    ///
-    /// This is currently also used for printing an expression as a match-arm,
-    /// but this is incorrect and leads to over-parenthesizing.
     pub fn new_stmt() -> Self {
         FixupContext { stmt: true, ..FixupContext::default() }
     }
 
+    /// Create the initial fixup for printing an expression as the right-hand
+    /// side of a match arm.
+    pub fn new_match_arm() -> Self {
+        FixupContext { match_arm: true, ..FixupContext::default() }
+    }
+
     /// Create the initial fixup for printing an expression as the "condition"
     /// of an `if` or `while`. There are a few other positions which are
     /// grammatically equivalent and also use this, such as the iterator
@@ -106,6 +143,9 @@ impl FixupContext {
         FixupContext {
             stmt: false,
             leftmost_subexpression_in_stmt: self.stmt || self.leftmost_subexpression_in_stmt,
+            match_arm: false,
+            leftmost_subexpression_in_match_arm: self.match_arm
+                || self.leftmost_subexpression_in_match_arm,
             ..self
         }
     }
@@ -119,7 +159,13 @@ impl FixupContext {
     /// example the `$b` in `$a + $b` and `-$b`, but not the one in `[$b]` or
     /// `$a.f($b)`.
     pub fn subsequent_subexpression(self) -> Self {
-        FixupContext { stmt: false, leftmost_subexpression_in_stmt: false, ..self }
+        FixupContext {
+            stmt: false,
+            leftmost_subexpression_in_stmt: false,
+            match_arm: false,
+            leftmost_subexpression_in_match_arm: false,
+            ..self
+        }
     }
 
     /// Determine whether parentheses are needed around the given expression to
@@ -128,7 +174,8 @@ impl FixupContext {
     /// The documentation on `FixupContext::leftmost_subexpression_in_stmt` has
     /// examples.
     pub fn would_cause_statement_boundary(self, expr: &Expr) -> bool {
-        self.leftmost_subexpression_in_stmt && !classify::expr_requires_semi_to_be_stmt(expr)
+        (self.leftmost_subexpression_in_stmt && !classify::expr_requires_semi_to_be_stmt(expr))
+            || (self.leftmost_subexpression_in_match_arm && classify::expr_is_complete(expr))
     }
 
     /// Determine whether parentheses are needed around the given `let`

Diff for: compiler/rustc_lint/src/unused.rs

@@ -677,6 +677,33 @@ trait UnusedDelimLint {
         }
 
         // Check if LHS needs parens to prevent false-positives in cases like `fn x() -> u8 { ({ 0 } + 1) }`.
+        //
+        // FIXME: https://github.com/rust-lang/rust/issues/119426
+        // The syntax tree in this code is from after macro expansion, so the
+        // current implementation has both false negatives and false positives
+        // related to expressions containing macros.
+        //
+        //     macro_rules! m1 {
+        //         () => {
+        //             1
+        //         };
+        //     }
+        //
+        //     fn f1() -> u8 {
+        //         // Lint says parens are not needed, but they are.
+        //         (m1! {} + 1)
+        //     }
+        //
+        //     macro_rules! m2 {
+        //         () => {
+        //             loop { break 1; }
+        //         };
+        //     }
+        //
+        //     fn f2() -> u8 {
+        //         // Lint says parens are needed, but they are not.
+        //         (m2!() + 1)
+        //     }
         {
             let mut innermost = inner;
             loop {

Diff for: compiler/rustc_parse/src/parser/expr.rs

@@ -497,8 +497,7 @@ impl<'a> Parser<'a> {
 
     /// Checks if this expression is a successfully parsed statement.
     fn expr_is_complete(&self, e: &Expr) -> bool {
-        self.restrictions.contains(Restrictions::STMT_EXPR)
-            && !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..=`.
@@ -2691,8 +2690,33 @@ impl<'a> Parser<'a> {
             let first_tok_span = self.token.span;
             match self.parse_expr() {
                 Ok(cond)
-                // If it's not a free-standing expression, and is followed by a block,
-                // then it's very likely the condition to an `else if`.
+                // Try to guess the difference between a "condition-like" vs
+                // "statement-like" expression.
+                //
+                // We are seeing the following code, in which $cond is neither
+                // ExprKind::Block nor ExprKind::If (the 2 cases wherein this
+                // would be valid syntax).
+                //
+                //     if ... {
+                //     } else $cond
+                //
+                // If $cond is "condition-like" such as ExprKind::Binary, we
+                // want to suggest inserting `if`.
+                //
+                //     if ... {
+                //     } else if a == b {
+                //            ^^
+                //     }
+                //
+                // If $cond is "statement-like" such as ExprKind::While then we
+                // want to suggest wrapping in braces.
+                //
+                //     if ... {
+                //     } else {
+                //            ^
+                //         while true {}
+                //     }
+                //     ^
                     if self.check(&TokenKind::OpenDelim(Delimiter::Brace))
                         && classify::expr_requires_semi_to_be_stmt(&cond) =>
                 {
@@ -3136,7 +3160,7 @@ impl<'a> Parser<'a> {
                         err
                     })?;
 
-                let require_comma = classify::expr_requires_semi_to_be_stmt(&expr)
+                let require_comma = !classify::expr_is_complete(&expr)
                     && this.token != token::CloseDelim(Delimiter::Brace);
 
                 if !require_comma {

Diff for: tests/ui/lint/lint-unnecessary-parens.fixed

@@ -46,6 +46,28 @@ pub fn parens_with_keyword(e: &[()]) -> i32 {
 macro_rules! baz {
     ($($foo:expr),+) => {
         ($($foo),*)
+    };
+}
+
+macro_rules! unit {
+    () => {
+        ()
+    };
+}
+
+struct One;
+
+impl std::ops::Sub<One> for () {
+    type Output = i32;
+    fn sub(self, _: One) -> Self::Output {
+        -1
+    }
+}
+
+impl std::ops::Neg for One {
+    type Output = i32;
+    fn neg(self) -> Self::Output {
+        -1
     }
 }
 
@@ -94,4 +116,13 @@ fn main() {
 
     let _a = baz!(3, 4);
     let _b = baz!(3);
+
+    let _ = {
+        unit!() - One //~ ERROR unnecessary parentheses around block return value
+    } + {
+        unit![] - One //~ ERROR unnecessary parentheses around block return value
+    } + {
+        // FIXME: false positive. This parenthesis is required.
+        unit! {} - One //~ ERROR unnecessary parentheses around block return value
+    };
 }

Diff for: tests/ui/lint/lint-unnecessary-parens.rs

@@ -46,6 +46,28 @@ pub fn parens_with_keyword(e: &[()]) -> i32 {
 macro_rules! baz {
     ($($foo:expr),+) => {
         ($($foo),*)
+    };
+}
+
+macro_rules! unit {
+    () => {
+        ()
+    };
+}
+
+struct One;
+
+impl std::ops::Sub<One> for () {
+    type Output = i32;
+    fn sub(self, _: One) -> Self::Output {
+        -1
+    }
+}
+
+impl std::ops::Neg for One {
+    type Output = i32;
+    fn neg(self) -> Self::Output {
+        -1
     }
 }
 
@@ -94,4 +116,13 @@ fn main() {
 
     let _a = baz!(3, 4);
     let _b = baz!(3);
+
+    let _ = {
+        (unit!() - One) //~ ERROR unnecessary parentheses around block return value
+    } + {
+        (unit![] - One) //~ ERROR unnecessary parentheses around block return value
+    } + {
+        // FIXME: false positive. This parenthesis is required.
+        (unit! {} - One) //~ ERROR unnecessary parentheses around block return value
+    };
 }