Rollup merge of rust-lang#62550 - Centril:rest-patterns, r=petrochenkov

Implement RFC 2707 + Parser recovery for range patterns

Implement rust-lang/rfcs#2707.

- Add a new basic syntactic pattern form `ast::PatKind::Rest` (parsed as `..` or `DOTDOT`) and simplify `ast::PatKind::{Slice, Tuple, TupleStruct}` as a result.

- Lower `ast::PatKind::Rest` in combination with the aforementioned `PatKind` variants as well as `PatKind::Ident`. The HIR remains unchanged for now (may be advisable to make slight adjustments later).

- Refactor `parser.rs` wrt. parsing sequences and lists of things in the process.

- Add parser recovery for range patterns of form `X..`, `X..=`, `X...`, `..Y`, `..=Y`, and `...Y`.
   This should make it easy to actually support these patterns semantically later if we so desire.

cc rust-lang#62254

r? @petrochenkov

Author: Centril

src/doc/unstable-book/src/language-features/slice-patterns.md

@@ -17,7 +17,7 @@ matched against that pattern. For example:
 fn is_symmetric(list: &[u32]) -> bool {
     match list {
         &[] | &[_] => true,
-        &[x, ref inside.., y] if x == y => is_symmetric(inside),
+        &[x, ref inside @ .., y] if x == y => is_symmetric(inside),
         &[..] => false,
     }
 }

src/librustc/hir/lowering.rs

@@ -58,6 +58,7 @@ use std::mem;
 use smallvec::SmallVec;
 use syntax::attr;
 use syntax::ast;
+use syntax::ptr::P as AstP;
 use syntax::ast::*;
 use syntax::errors;
 use syntax::ext::hygiene::ExpnId;
@@ -467,7 +468,7 @@ impl<'a> LoweringContext<'a> {
             fn visit_pat(&mut self, p: &'tcx Pat) {
                 match p.node {
                     // Doesn't generate a HIR node
-                    PatKind::Paren(..) => {},
+                    PatKind::Paren(..) | PatKind::Rest => {},
                     _ => {
                         if let Some(owner) = self.hir_id_owner {
                             self.lctx.lower_node_id_with_owner(p.id, owner);
@@ -1156,7 +1157,7 @@ impl<'a> LoweringContext<'a> {
         &mut self,
         capture_clause: CaptureBy,
         closure_node_id: NodeId,
-        ret_ty: Option<syntax::ptr::P<Ty>>,
+        ret_ty: Option<AstP<Ty>>,
         span: Span,
         body: impl FnOnce(&mut LoweringContext<'_>) -> hir::Expr,
     ) -> hir::ExprKind {
@@ -4171,45 +4172,20 @@ impl<'a> LoweringContext<'a> {
         let node = match p.node {
             PatKind::Wild => hir::PatKind::Wild,
             PatKind::Ident(ref binding_mode, ident, ref sub) => {
-                match self.resolver.get_partial_res(p.id).map(|d| d.base_res()) {
-                    // `None` can occur in body-less function signatures
-                    res @ None | res @ Some(Res::Local(_)) => {
-                        let canonical_id = match res {
-                            Some(Res::Local(id)) => id,
-                            _ => p.id,
-                        };
-
-                        hir::PatKind::Binding(
-                            self.lower_binding_mode(binding_mode),
-                            self.lower_node_id(canonical_id),
-                            ident,
-                            sub.as_ref().map(|x| self.lower_pat(x)),
-                        )
-                    }
-                    Some(res) => hir::PatKind::Path(hir::QPath::Resolved(
-                        None,
-                        P(hir::Path {
-                            span: ident.span,
-                            res: self.lower_res(res),
-                            segments: hir_vec![hir::PathSegment::from_ident(ident)],
-                        }),
-                    )),
-                }
+                let lower_sub = |this: &mut Self| sub.as_ref().map(|x| this.lower_pat(x));
+                self.lower_pat_ident(p, binding_mode, ident, lower_sub)
             }
             PatKind::Lit(ref e) => hir::PatKind::Lit(P(self.lower_expr(e))),
-            PatKind::TupleStruct(ref path, ref pats, ddpos) => {
+            PatKind::TupleStruct(ref path, ref pats) => {
                 let qpath = self.lower_qpath(
                     p.id,
                     &None,
                     path,
                     ParamMode::Optional,
                     ImplTraitContext::disallowed(),
                 );
-                hir::PatKind::TupleStruct(
-                    qpath,
-                    pats.iter().map(|x| self.lower_pat(x)).collect(),
-                    ddpos,
-                )
+                let (pats, ddpos) = self.lower_pat_tuple(pats, "tuple struct");
+                hir::PatKind::TupleStruct(qpath, pats, ddpos)
             }
             PatKind::Path(ref qself, ref path) => {
                 let qpath = self.lower_qpath(
@@ -4246,8 +4222,9 @@ impl<'a> LoweringContext<'a> {
                     .collect();
                 hir::PatKind::Struct(qpath, fs, etc)
             }
-            PatKind::Tuple(ref elts, ddpos) => {
-                hir::PatKind::Tuple(elts.iter().map(|x| self.lower_pat(x)).collect(), ddpos)
+            PatKind::Tuple(ref pats) => {
+                let (pats, ddpos) = self.lower_pat_tuple(pats, "tuple");
+                hir::PatKind::Tuple(pats, ddpos)
             }
             PatKind::Box(ref inner) => hir::PatKind::Box(self.lower_pat(inner)),
             PatKind::Ref(ref inner, mutbl) => {
@@ -4258,22 +4235,167 @@ impl<'a> LoweringContext<'a> {
                 P(self.lower_expr(e2)),
                 self.lower_range_end(end),
             ),
-            PatKind::Slice(ref before, ref slice, ref after) => hir::PatKind::Slice(
-                before.iter().map(|x| self.lower_pat(x)).collect(),
-                slice.as_ref().map(|x| self.lower_pat(x)),
-                after.iter().map(|x| self.lower_pat(x)).collect(),
-            ),
+            PatKind::Slice(ref pats) => self.lower_pat_slice(pats),
+            PatKind::Rest => {
+                // If we reach here the `..` pattern is not semantically allowed.
+                self.ban_illegal_rest_pat(p.span)
+            }
             PatKind::Paren(ref inner) => return self.lower_pat(inner),
             PatKind::Mac(_) => panic!("Shouldn't exist here"),
         };
 
+        self.pat_with_node_id_of(p, node)
+    }
+
+    fn lower_pat_tuple(
+        &mut self,
+        pats: &[AstP<Pat>],
+        ctx: &str,
+    ) -> (HirVec<P<hir::Pat>>, Option<usize>) {
+        let mut elems = Vec::with_capacity(pats.len());
+        let mut rest = None;
+
+        let mut iter = pats.iter().enumerate();
+        while let Some((idx, pat)) = iter.next() {
+            // Interpret the first `..` pattern as a subtuple pattern.
+            if pat.is_rest() {
+                rest = Some((idx, pat.span));
+                break;
+            }
+            // It was not a subslice pattern so lower it normally.
+            elems.push(self.lower_pat(pat));
+        }
+
+        while let Some((_, pat)) = iter.next() {
+            // There was a previous subtuple pattern; make sure we don't allow more.
+            if pat.is_rest() {
+                self.ban_extra_rest_pat(pat.span, rest.unwrap().1, ctx);
+            } else {
+                elems.push(self.lower_pat(pat));
+            }
+        }
+
+        (elems.into(), rest.map(|(ddpos, _)| ddpos))
+    }
+
+    fn lower_pat_slice(&mut self, pats: &[AstP<Pat>]) -> hir::PatKind {
+        let mut before = Vec::new();
+        let mut after = Vec::new();
+        let mut slice = None;
+        let mut prev_rest_span = None;
+
+        let mut iter = pats.iter();
+        while let Some(pat) = iter.next() {
+            // Interpret the first `((ref mut?)? x @)? ..` pattern as a subslice pattern.
+            match pat.node {
+                PatKind::Rest => {
+                    prev_rest_span = Some(pat.span);
+                    slice = Some(self.pat_wild_with_node_id_of(pat));
+                    break;
+                },
+                PatKind::Ident(ref bm, ident, Some(ref sub)) if sub.is_rest() => {
+                    prev_rest_span = Some(sub.span);
+                    let lower_sub = |this: &mut Self| Some(this.pat_wild_with_node_id_of(sub));
+                    let node = self.lower_pat_ident(pat, bm, ident, lower_sub);
+                    slice = Some(self.pat_with_node_id_of(pat, node));
+                    break;
+                },
+                _ => {}
+            }
+
+            // It was not a subslice pattern so lower it normally.
+            before.push(self.lower_pat(pat));
+        }
+
+        while let Some(pat) = iter.next() {
+            // There was a previous subslice pattern; make sure we don't allow more.
+            let rest_span = match pat.node {
+                PatKind::Rest => Some(pat.span),
+                PatKind::Ident(.., Some(ref sub)) if sub.is_rest() => {
+                    // The `HirValidator` is merciless; add a `_` pattern to avoid ICEs.
+                    after.push(self.pat_wild_with_node_id_of(pat));
+                    Some(sub.span)
+                },
+                _ => None,
+            };
+            if let Some(rest_span) = rest_span {
+                self.ban_extra_rest_pat(rest_span, prev_rest_span.unwrap(), "slice");
+            } else {
+                after.push(self.lower_pat(pat));
+            }
+        }
+
+        hir::PatKind::Slice(before.into(), slice, after.into())
+    }
+
+    fn lower_pat_ident(
+        &mut self,
+        p: &Pat,
+        binding_mode: &BindingMode,
+        ident: Ident,
+        lower_sub: impl FnOnce(&mut Self) -> Option<P<hir::Pat>>,
+    ) -> hir::PatKind {
+        match self.resolver.get_partial_res(p.id).map(|d| d.base_res()) {
+            // `None` can occur in body-less function signatures
+            res @ None | res @ Some(Res::Local(_)) => {
+                let canonical_id = match res {
+                    Some(Res::Local(id)) => id,
+                    _ => p.id,
+                };
+
+                hir::PatKind::Binding(
+                    self.lower_binding_mode(binding_mode),
+                    self.lower_node_id(canonical_id),
+                    ident,
+                    lower_sub(self),
+                )
+            }
+            Some(res) => hir::PatKind::Path(hir::QPath::Resolved(
+                None,
+                P(hir::Path {
+                    span: ident.span,
+                    res: self.lower_res(res),
+                    segments: hir_vec![hir::PathSegment::from_ident(ident)],
+                }),
+            )),
+        }
+    }
+
+    fn pat_wild_with_node_id_of(&mut self, p: &Pat) -> P<hir::Pat> {
+        self.pat_with_node_id_of(p, hir::PatKind::Wild)
+    }
+
+    /// Construct a `Pat` with the `HirId` of `p.id` lowered.
+    fn pat_with_node_id_of(&mut self, p: &Pat, node: hir::PatKind) -> P<hir::Pat> {
         P(hir::Pat {
             hir_id: self.lower_node_id(p.id),
             node,
             span: p.span,
         })
     }
 
+    /// Emit a friendly error for extra `..` patterns in a tuple/tuple struct/slice pattern.
+    fn ban_extra_rest_pat(&self, sp: Span, prev_sp: Span, ctx: &str) {
+        self.diagnostic()
+            .struct_span_err(sp, &format!("`..` can only be used once per {} pattern", ctx))
+            .span_label(sp, &format!("can only be used once per {} pattern", ctx))
+            .span_label(prev_sp, "previously used here")
+            .emit();
+    }
+
+    /// Used to ban the `..` pattern in places it shouldn't be semantically.
+    fn ban_illegal_rest_pat(&self, sp: Span) -> hir::PatKind {
+        self.diagnostic()
+            .struct_span_err(sp, "`..` patterns are not allowed here")
+            .note("only allowed in tuple, tuple struct, and slice patterns")
+            .emit();
+
+        // We're not in a list context so `..` can be reasonably treated
+        // as `_` because it should always be valid and roughly matches the
+        // intent of `..` (notice that the rest of a single slot is that slot).
+        hir::PatKind::Wild
+    }
+
     fn lower_range_end(&mut self, e: &RangeEnd) -> hir::RangeEnd {
         match *e {
             RangeEnd::Included(_) => hir::RangeEnd::Included,

src/librustc_mir/hair/pattern/mod.rs

@@ -162,7 +162,7 @@ pub enum PatternKind<'tcx> {
 
     /// Matches against a slice, checking the length and extracting elements.
     /// irrefutable when there is a slice pattern and both `prefix` and `suffix` are empty.
-    /// e.g., `&[ref xs..]`.
+    /// e.g., `&[ref xs @ ..]`.
     Slice {
         prefix: Vec<Pattern<'tcx>>,
         slice: Option<Pattern<'tcx>>,

src/librustc_passes/ast_validation.rs

@@ -287,7 +287,7 @@ impl<'a> AstValidator<'a> {
     // ```
     fn check_expr_within_pat(&self, expr: &Expr, allow_paths: bool) {
         match expr.node {
-            ExprKind::Lit(..) => {}
+            ExprKind::Lit(..) | ExprKind::Err => {}
             ExprKind::Path(..) if allow_paths => {}
             ExprKind::Unary(UnOp::Neg, ref inner)
                 if match inner.node { ExprKind::Lit(_) => true, _ => false } => {}

src/librustc_target/abi/mod.rs

@@ -105,20 +105,34 @@ impl TargetDataLayout {
         let mut dl = TargetDataLayout::default();
         let mut i128_align_src = 64;
         for spec in target.data_layout.split('-') {
-            match spec.split(':').collect::<Vec<_>>()[..] {
+            let spec_parts = spec.split(':').collect::<Vec<_>>();
+
+            match &*spec_parts {
                 ["e"] => dl.endian = Endian::Little,
                 ["E"] => dl.endian = Endian::Big,
                 [p] if p.starts_with("P") => {
                     dl.instruction_address_space = parse_address_space(&p[1..], "P")?
                 }
-                ["a", ref a..] => dl.aggregate_align = align(a, "a")?,
-                ["f32", ref a..] => dl.f32_align = align(a, "f32")?,
-                ["f64", ref a..] => dl.f64_align = align(a, "f64")?,
-                [p @ "p", s, ref a..] | [p @ "p0", s, ref a..] => {
+                // FIXME: Ping cfg(bootstrap) -- Use `ref a @ ..` with new bootstrap compiler.
+                ["a", ..] => {
+                    let a = &spec_parts[1..]; // FIXME inline into pattern.
+                    dl.aggregate_align = align(a, "a")?
+                }
+                ["f32", ..] => {
+                    let a = &spec_parts[1..]; // FIXME inline into pattern.
+                    dl.f32_align = align(a, "f32")?
+                }
+                ["f64", ..] => {
+                    let a = &spec_parts[1..]; // FIXME inline into pattern.
+                    dl.f64_align = align(a, "f64")?
+                }
+                [p @ "p", s, ..] | [p @ "p0", s, ..] => {
+                    let a = &spec_parts[2..]; // FIXME inline into pattern.
                     dl.pointer_size = size(s, p)?;
                     dl.pointer_align = align(a, p)?;
                 }
-                [s, ref a..] if s.starts_with("i") => {
+                [s, ..] if s.starts_with("i") => {
+                    let a = &spec_parts[1..]; // FIXME inline into pattern.
                     let bits = match s[1..].parse::<u64>() {
                         Ok(bits) => bits,
                         Err(_) => {
@@ -142,7 +156,8 @@ impl TargetDataLayout {
                         dl.i128_align = a;
                     }
                 }
-                [s, ref a..] if s.starts_with("v") => {
+                [s, ..] if s.starts_with("v") => {
+                    let a = &spec_parts[1..]; // FIXME inline into pattern.
                     let v_size = size(&s[1..], "v")?;
                     let a = align(a, s)?;
                     if let Some(v) = dl.vector_align.iter_mut().find(|v| v.0 == v_size) {

src/librustc_typeck/check/_match.rs

@@ -196,7 +196,11 @@ impl<'a, 'tcx> FnCtxt<'a, 'tcx> {
                 let rhs_ty = self.check_expr(end);
 
                 // Check that both end-points are of numeric or char type.
-                let numeric_or_char = |ty: Ty<'_>| ty.is_numeric() || ty.is_char();
+                let numeric_or_char = |ty: Ty<'_>| {
+                    ty.is_numeric()
+                    || ty.is_char()
+                    || ty.references_error()
+                };
                 let lhs_compat = numeric_or_char(lhs_ty);
                 let rhs_compat = numeric_or_char(rhs_ty);
 

src/librustc_typeck/check/mod.rs

@@ -1818,7 +1818,9 @@ fn bad_variant_count<'tcx>(tcx: TyCtxt<'tcx>, adt: &'tcx ty::AdtDef, sp: Span, d
     );
     let mut err = struct_span_err!(tcx.sess, sp, E0731, "transparent enum {}", msg);
     err.span_label(sp, &msg);
-    if let &[ref start.., ref end] = &variant_spans[..] {
+    if let &[.., ref end] = &variant_spans[..] {
+        // FIXME: Ping cfg(bootstrap) -- Use `ref start @ ..` with new bootstrap compiler.
+        let start = &variant_spans[..variant_spans.len() - 1];
         for variant_span in start {
             err.span_label(*variant_span, "");
         }

src/librustc_typeck/error_codes.rs

@@ -3497,8 +3497,8 @@ Example of erroneous code:
 
 let r = &[1, 2];
 match r {
-    &[a, b, c, rest..] => { // error: pattern requires at least 3
-                            //        elements but array has 2
+    &[a, b, c, rest @ ..] => { // error: pattern requires at least 3
+                               //        elements but array has 2
         println!("a={}, b={}, c={} rest={:?}", a, b, c, rest);
     }
 }
@@ -3512,7 +3512,7 @@ requires. You can match an arbitrary number of remaining elements with `..`:
 
 let r = &[1, 2, 3, 4, 5];
 match r {
-    &[a, b, c, rest..] => { // ok!
+    &[a, b, c, rest @ ..] => { // ok!
         // prints `a=1, b=2, c=3 rest=[4, 5]`
         println!("a={}, b={}, c={} rest={:?}", a, b, c, rest);
     }