fix: 同名捕获组回引与 groups 字段改末次已定义语义

2026-10-23 18:21:13 +08:00 · 2026-10-14 08:05:13 +00:00
parent 3d3bdaa5cf
commit fe430d4bcb
12 changed files with 287 additions and 42 deletions
@@ -1160,7 +1160,7 @@ dependencies = [

 [[package]]
 name = "oxide_builtins"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "chrono",
 "fast-float",
@@ -1185,7 +1185,7 @@ dependencies = [

 [[package]]
 name = "oxide_bytecode"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "num-bigint",
 "oxide_log",
@@ -1193,7 +1193,7 @@ dependencies = [

 [[package]]
 name = "oxide_cfg"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_emit",
@@ -1204,7 +1204,7 @@ dependencies = [

 [[package]]
 name = "oxide_cli"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "ansi_term",
 "bumpalo",
@@ -1228,7 +1228,7 @@ dependencies = [

 [[package]]
 name = "oxide_code_cache"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "lru",
 "oxide_bytecode",
@@ -1237,7 +1237,7 @@ dependencies = [

 [[package]]
 name = "oxide_compiler"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_cfg",
@@ -1258,7 +1258,7 @@ dependencies = [

 [[package]]
 name = "oxide_dce"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_cfg",
@@ -1272,7 +1272,7 @@ dependencies = [

 [[package]]
 name = "oxide_emit"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "num-bigint",
 "oxc_allocator",
@@ -1288,7 +1288,7 @@ dependencies = [

 [[package]]
 name = "oxide_engine"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_compiler",
@@ -1300,7 +1300,7 @@ dependencies = [

 [[package]]
 name = "oxide_fold"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "num-bigint",
 "oxide_bytecode",
@@ -1312,7 +1312,7 @@ dependencies = [

 [[package]]
 name = "oxide_ir"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_log",
@@ -1321,7 +1321,7 @@ dependencies = [

 [[package]]
 name = "oxide_kernel"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "bumpalo",
 "dashmap",
@@ -1335,7 +1335,7 @@ dependencies = [

 [[package]]
 name = "oxide_liveness"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_cfg",
@@ -1347,7 +1347,7 @@ dependencies = [

 [[package]]
 name = "oxide_log"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "tracing",
 "tracing-appender",
@@ -1356,11 +1356,11 @@ dependencies = [

 [[package]]
 name = "oxide_net"
-version = "0.19.7"
+version = "0.19.8"

 [[package]]
 name = "oxide_parser"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxc_allocator",
 "oxc_ast",
@@ -1375,7 +1375,7 @@ dependencies = [

 [[package]]
 name = "oxide_regalloc"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_cfg",
@@ -1391,7 +1391,7 @@ dependencies = [

 [[package]]
 name = "oxide_runtime_api"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "num-bigint",
 "num-traits",
@@ -1404,7 +1404,7 @@ dependencies = [

 [[package]]
 name = "oxide_test262"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "oxide_bytecode",
 "oxide_compiler",
@@ -1424,7 +1424,7 @@ dependencies = [

 [[package]]
 name = "oxide_types"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "bumpalo",
 "dashmap",
@@ -1437,7 +1437,7 @@ dependencies = [

 [[package]]
 name = "oxide_vm"
-version = "0.19.7"
+version = "0.19.8"
 dependencies = [
 "bumpalo",
 "chrono",
@@ -25,7 +25,7 @@ members = [
 ]

 [workspace.package]
-version = "0.19.7"
+version = "0.19.8"
 edition = "2021"
 license = "MIT"
 rust-version = "1.80"
@@ -18,7 +18,7 @@ pub(crate) fn build_groups_object<H: VmHost>(vm: &mut H, m: &regress::Match, tex
    let groups_obj = vm.alloc_object(JsObject::new_empty(EMPTY_SHAPE_ID, JsValue::null()));
    let groups_ptr = unsafe { &mut *groups_obj };

-    // 遍历所有命名捕获组，按名称设置属性（重复名称取首个已定义的出现）。
+    // 遍历所有命名捕获组，按名称设置属性（重复名称取末次已定义的出现）。
    for (name, range) in named {
        let name_si = vm.perm_intern(name);
        let value = match range {
@@ -2222,3 +2222,78 @@ pub fn regexp_symbol_match_all<H: VmHost>(vm: &mut H, args: &[u8]) -> NativeResu
    vm.set_or_create_prop_value(wrapper_obj, unicode_si, JsValue::bool(is_unicode));
    NativeResult::Ok(JsValue::from_js_object(wrapper))
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::sync::Arc;
+
+    use oxide_vm::vm::Vm;
+
+    /// 编译并执行单脚本，返回 VM 与完成值。
+    fn eval(source: &str) -> Result<(Vm, JsValue), String> {
+        let allocator = oxide_parser::Allocator::default();
+        let program = oxide_parser::parse(&allocator, source).map_err(|e| format!("parse: {:?}", e))?;
+        let module = oxide_compiler::compiler::Compiler::new()
+            .compile(&program)
+            .map_err(|e| format!("compile: {e}"))?;
+        let mut vm = Vm::new();
+        let result = vm.run(&Arc::new(module))?;
+        Ok((vm, result))
+    }
+
+    /// 读对象自身属性值（按名称），无则 undefined。
+    fn obj_prop(vm: &Vm, obj: &JsObject, name: &str) -> JsValue {
+        let si = vm.perm_intern(name);
+        match vm.get_own_property_slot(obj, si) {
+            Some(idx) => obj.get_prop_at(idx),
+            None => JsValue::undefined(),
+        }
+    }
+
+    /// 读 exec 结果数组第 i 个元素的字符串值（按值）。
+    fn arr_str(vm: &Vm, arr: &JsObject, i: usize) -> Option<String> {
+        vm.lookup_str(arr.get_prop_at(i))
+    }
+
+    /// groups.x 取末次已定义出现；result 数组各出现独立交付（首出现被迭代 2
+    /// 重入清槽，末次出现命中 b）。
+    #[test]
+    fn groups_last_defined() {
+        let (vm, result) = eval(r"/(?:(?<x>a)|(?<x>b)){2}/.exec('ab')").unwrap();
+        let arr = unsafe { &*result.as_js_object_ptr() };
+        assert_eq!(arr_str(&vm, arr, 0).as_deref(), Some("ab"));
+        assert!(arr.get_prop_at(1).is_undefined());
+        assert_eq!(arr_str(&vm, arr, 2).as_deref(), Some("b"));
+        let groups = obj_prop(&vm, arr, "groups");
+        let groups_obj = unsafe { &*groups.as_js_object_ptr() };
+        assert_eq!(vm.lookup_str(obj_prop(&vm, groups_obj, "x")).as_deref(), Some("b"));
+    }
+
+    /// 同名回引读末次已定义：迭代 2 中末次出现命中 b，回引读 b 命中 "aabb"。
+    #[test]
+    fn backref_last_defined_aabb() {
+        let (vm, result) = eval(r"/(?:(?:(?<x>a)|(?<x>b))\k<x>){2}/.exec('aabb')").unwrap();
+        let arr = unsafe { &*result.as_js_object_ptr() };
+        assert_eq!(arr_str(&vm, arr, 0).as_deref(), Some("aabb"));
+        assert!(arr.get_prop_at(1).is_undefined(), "首出现重入未命中");
+        assert_eq!(arr_str(&vm, arr, 2).as_deref(), Some("b"));
+        let groups = obj_prop(&vm, arr, "groups");
+        let groups_obj = unsafe { &*groups.as_js_object_ptr() };
+        assert_eq!(vm.lookup_str(obj_prop(&vm, groups_obj, "x")).as_deref(), Some("b"));
+    }
+
+    /// 全部出现未定义时回引空匹配；末次已定义值不匹配时整体失败（不回落空匹配）。
+    #[test]
+    fn backref_undefined_empty_and_no_fallthrough() {
+        let (vm, result) = eval(r"/^(?:(?<a>x)|(?<a>y)|z)\k<a>$/.exec('z')").unwrap();
+        let arr = unsafe { &*result.as_js_object_ptr() };
+        assert_eq!(arr_str(&vm, arr, 0).as_deref(), Some("z"));
+        assert!(arr.get_prop_at(1).is_undefined());
+        assert!(arr.get_prop_at(2).is_undefined());
+
+        // 末次已定义为 x（位 2 为 y，不匹配 x），整体应失败。
+        let (_, result) = eval(r"/(?:(?<a>y)|(?<a>x)){2}\k<a>/.exec('yxy')").unwrap();
+        assert!(result.is_null(), "末次已定义值不匹配时整体应失败");
+    }
+}
@@ -339,15 +339,14 @@ impl<'m> Iterator for NamedGroups<'m> {
                continue;
            }

-            // This is the first occurrence of this name. Find the best range value.
-            // Prefer a Some value over None when there are duplicate names.
+            // This is the first occurrence of this name. Resolve the value to
+            // the last occurrence of this name that matched: a later duplicate
+            // overrides an earlier one only when it is defined.
            let mut best_range = self.mat.captures[idx].clone();
            for check_idx in (idx + 1)..end {
                if self.mat.group_names[check_idx].as_ref() == name {
-                    // Found a duplicate name. Prefer a Some value over None.
-                    if best_range.is_none() && self.mat.captures[check_idx].is_some() {
-                        best_range = self.mat.captures[check_idx].clone();
-                        break; // Stop once we find a Some value
+                    if let Some(range) = self.mat.captures[check_idx].as_ref() {
+                        best_range = Some(range.clone());
                    }
                }
            }
@@ -1272,3 +1271,112 @@ mod utf16_tests {
        assert_eq!(str_matches(&re, text), utf16_matches(&re, &to_units(text)));
    }
 }
+
+#[cfg(test)]
+mod named_groups_tests {
+    use super::Regex;
+
+    /// 同名双定义（循环内各迭代命中不同出现）：named_groups 取末次已定义出现，
+    /// 键序为首现源序。
+    #[test]
+    fn named_groups_last_defined_wins() {
+        let re = Regex::new(r"(?:(?<x>a)|(?<x>b)){2}").unwrap();
+        let m = re.find("ab").unwrap();
+        let named: Vec<_> = m.named_groups().collect();
+        assert_eq!(named, vec![("x", Some(1..2))], "末次已定义出现（b）胜出");
+        // 首出现被迭代 2 重入清槽，末次出现命中 b。
+        assert_eq!(m.captures, vec![None, Some(1..2)], "result 数组各出现独立交付");
+    }
+
+    /// 末次出现未定义时回退至更早已定义出现。
+    #[test]
+    fn named_groups_falls_back_to_earlier_defined() {
+        let re = Regex::new(r"(?:(?<x>a)|(?<x>b)|c){2}").unwrap();
+        let m = re.find("ca").unwrap();
+        let named: Vec<_> = m.named_groups().collect();
+        assert_eq!(named, vec![("x", Some(1..2))], "末次出现未定义，回退首出现");
+        assert_eq!(m.captures, vec![Some(1..2), None]);
+    }
+
+    /// 全部出现未定义时取 undefined。
+    #[test]
+    fn named_groups_all_undefined() {
+        let re = Regex::new(r"(?:(?<x>a)|(?<x>b)|c){2}").unwrap();
+        let m = re.find("cc").unwrap();
+        let named: Vec<_> = m.named_groups().collect();
+        assert_eq!(named, vec![("x", None)]);
+        assert_eq!(m.captures, vec![None, None]);
+    }
+
+    /// 同名回引读末次已定义出现：迭代 2 中末次出现命中 b，回引读 b 命中 "aabb"。
+    #[test]
+    fn backref_reads_last_defined() {
+        let re = Regex::new(r"(?:(?:(?<x>a)|(?<x>b))\k<x>){2}").unwrap();
+        let m = re.find("aabb").unwrap();
+        assert_eq!(m.range(), 0..4);
+        assert_eq!(m.captures, vec![None, Some(2..3)], "各出现独立交付");
+    }
+
+    /// 单轮同名回引：末次已定义分别为首出现（aa）与次出现（bb）。
+    #[test]
+    fn backref_single_round() {
+        let re = Regex::new(r"(?:(?<x>a)|(?<x>b))\k<x>").unwrap();
+        let m = re.find("aa").unwrap();
+        assert_eq!(m.range(), 0..2);
+        assert_eq!(m.captures, vec![Some(0..1), None]);
+        let m = re.find("bb").unwrap();
+        assert_eq!(m.range(), 0..2);
+        assert_eq!(m.captures, vec![None, Some(0..1)]);
+    }
+
+    /// 全部出现未定义时回引空匹配；末次已定义值不匹配时整体失败（不回落空匹配，
+    /// 也不回落更早已定义出现）。
+    #[test]
+    fn backref_undefined_empty_and_no_fallthrough() {
+        let re = Regex::new(r"^(?:(?<a>x)|(?<a>y)|z)\k<a>$").unwrap();
+        let m = re.find("z").unwrap();
+        assert_eq!(m.range(), 0..1);
+        assert_eq!(m.captures, vec![None, None]);
+        assert!(re.find("zz").is_none(), "末次已定义值不匹配时整体失败");
+
+        // 末次已定义为 x（位 2 为 y，不匹配 x）；更早已定义为 y（位 2 亦不匹配）。
+        let re = Regex::new(r"(?:(?<a>y)|(?<a>x)){2}\k<a>").unwrap();
+        assert!(re.find("yxy").is_none(), "末次已定义值不匹配时整体失败");
+        // 对照：末次已定义值匹配时命中。
+        let m = re.find("yxx").unwrap();
+        assert_eq!(m.range(), 0..3);
+    }
+
+    /// 循环重入清槽后全部未定义：回引空匹配，结果数组各出现 undefined。
+    #[test]
+    fn backref_loop_reentry_clears() {
+        let re = Regex::new(r"^(?:(?<a>x)|(?<a>y)|z){2}\k<a>$").unwrap();
+        let m = re.find("xz").unwrap();
+        assert_eq!(m.range(), 0..2);
+        assert_eq!(m.captures, vec![None, None]);
+        assert!(re.find("xzx").is_none());
+    }
+
+    /// 未定义组回引空匹配（单出现面不变）。
+    #[test]
+    fn backref_single_occurrence_undefined_empty() {
+        let re = Regex::new(r"(?<a>x)|(?:zy\k<a>)").unwrap();
+        let m = re.find("zy").unwrap();
+        assert_eq!(m.range(), 0..2);
+        assert_eq!(m.captures, vec![None]);
+    }
+
+    /// PikeVM 执行器与回退执行器对同名回引口径一致。
+    #[cfg(feature = "backend-pikevm")]
+    #[test]
+    fn pikevm_backref_agrees() {
+        use crate::backends::PikeVMExecutor;
+        use crate::exec::{Executor, Matches};
+        let re = Regex::new(r"(?:(?:(?<x>a)|(?<x>b))\k<x>){2}").unwrap();
+        let m = Matches::new(PikeVMExecutor::new(&re.cr, "aabb"), 0)
+            .next()
+            .expect("pikevm 应命中 aabb");
+        assert_eq!(m.range(), 0..4);
+        assert_eq!(m.captures, vec![None, Some(2..3)]);
+    }
+}
@@ -886,6 +886,27 @@ impl<'a, Input: InputIndexer> MatchAttempter<'a, Input> {
                        next_or_bt!(matched)
                    }

+                    Insn::BackRefLastDefined(cg_indices) => {
+                        // Scan the occurrences in reverse source order and match the
+                        // text of the last occurrence that is defined; a later
+                        // occurrence overrides an earlier one only when it matched.
+                        let mut matched = true;
+                        for &cg_idx in cg_indices.iter().rev() {
+                            let cg = self.s.groups.mat(cg_idx as usize);
+                            if let Some(orig_range) = cg.as_range() {
+                                if re.flags.icase {
+                                    matched = matchers::backref_icase(input, dir, orig_range, &mut pos);
+                                } else {
+                                    matched = matchers::backref(input, dir, orig_range, &mut pos);
+                                }
+                                break;
+                            }
+                        }
+                        // No occurrence is defined and so the match succeeds (ES6
+                        // 21.2.2.9).
+                        next_or_bt!(matched)
+                    }
+
                    &Insn::Lookahead {
                        negate,
                        start_group,
@@ -340,6 +340,14 @@ impl Emitter {
                        self.emit_insn(Insn::BackRef(group - 1))
                    }

+                    Node::BackRefLastDefined(groups) => {
+                        debug_assert!(!groups.is_empty(), "Should not be empty");
+                        // -1 because group indices are 1-based in the IR.
+                        let indices: Vec<CaptureGroupID> =
+                            groups.iter().map(|g| (*g - 1) as CaptureGroupID).collect();
+                        self.emit_insn(Insn::BackRefLastDefined(indices))
+                    }
+
                    Node::ByteSet(bytes) => self.emit_byte_set_insn(bytes),

                    Node::CharSet(chars) => {
@@ -102,6 +102,11 @@ pub enum Insn {
    /// Perform a backreference match.
    BackRef(u32),

+    /// Perform a backreference match against the last-defined occurrence of a
+    /// name shared by multiple capture groups. Indices are 0-based, in source
+    /// order; if no occurrence is defined the match succeeds.
+    BackRefLastDefined(Vec<CaptureGroupID>),
+
    /// Match the next character against the bracket contents, stored at the given index in the CompiledRegex.
    Bracket(usize),

@@ -84,10 +84,14 @@ pub enum Node {
    /// A backreference.
    /// Indices are 1-based. That is, it matches JS syntax: \1 is the first capture group,
    /// not everything.
-    /// A backreference that logically may match multiple capture groups (through shared names)
-    /// are emitted through alternations of backreferences.
    BackRef(u32),

+    /// A backreference to a name shared by multiple capture groups.
+    /// Indices are 1-based, one per occurrence in source order. At match time it
+    /// matches the text of the last occurrence that is defined; if none is
+    /// defined, it matches the empty string.
+    BackRefLastDefined(Vec<u32>),
+
    /// A bracket.
    Bracket(BracketContents),

@@ -232,6 +236,7 @@ impl Node {
            }
            &Node::WordBoundary { invert } => Node::WordBoundary { invert },
            &Node::BackRef(idx) => Node::BackRef(idx),
+            Node::BackRefLastDefined(groups) => Node::BackRefLastDefined(groups.clone()),
            Node::Bracket(bc) => Node::Bracket(bc.clone()),
            // Do not reverse into lookarounds, they already have the right sense.
            Node::LookaroundAssertion {
@@ -308,6 +313,7 @@ where
            | Node::CharSet(..)
            | Node::WordBoundary { .. }
            | Node::BackRef { .. }
+            | Node::BackRefLastDefined { .. }
            | Node::Bracket { .. }
            | Node::MatchAny
            | Node::MatchAnyExceptLineTerminator
@@ -381,6 +387,7 @@ where
            | Node::Anchor { .. }
            | Node::WordBoundary { .. }
            | Node::BackRef { .. }
+            | Node::BackRefLastDefined { .. }
            | Node::Bracket { .. } => {}
            Node::Cat(nodes) => {
                nodes.iter_mut().for_each(|node| self.process(node));
@@ -548,6 +555,9 @@ fn display_node(node: &Node, depth: usize, f: &mut fmt::Formatter) -> fmt::Resul
        &Node::BackRef(group) => {
            writeln!(f, "BackRef {:?} ", group)?;
        }
+        Node::BackRefLastDefined(groups) => {
+            writeln!(f, "BackRefLastDefined {:?} ", groups)?;
+        }
        Node::Bracket(contents) => {
            writeln!(f, "Bracket {:?}", contents)?;
        }
@@ -160,7 +160,10 @@ fn remove_empties(n: &mut Node, _w: &Walk) -> PassAction {
            // Capture groups could in principle be optimized if they only match empties.
            PassAction::Keep
        }
-        Node::WordBoundary { .. } | Node::BackRef { .. } | Node::Bracket { .. } => PassAction::Keep,
+        Node::WordBoundary { .. }
+        | Node::BackRef { .. }
+        | Node::BackRefLastDefined { .. }
+        | Node::Bracket { .. } => PassAction::Keep,
        Node::LookaroundAssertion {
            negate, contents, ..
        } => {
@@ -1689,15 +1689,9 @@ where
                        ir::Node::BackRef(group_indices[0] + 1)
                    }
                    _ => {
-                        // Unusual case of multiple groups sharing a name: the backref should try each in turn.
-                        // Lower to alternations of backreferences. Reverse to keep it right-associative: a | (b | (c | d))...
-                        let backrefs = group_indices
-                            .iter()
-                            .rev()
-                            .map(|group_index| ir::Node::BackRef(*group_index + 1));
-                        backrefs
-                            .reduce(|right, left| ir::Node::Alt(Box::new(left), Box::new(right)))
-                            .unwrap()
+                        // Multiple groups sharing a name: the backref reads the
+                        // last occurrence that is defined at match time.
+                        ir::Node::BackRefLastDefined(group_indices.iter().map(|i| i + 1).collect())
                    }
                };
                Ok(node)
@@ -238,6 +238,27 @@ fn try_match_state<Input: InputIndexer, Dir: Direction>(
            nextinsn_or_fail!(matched)
        }

+        Insn::BackRefLastDefined(group_indices) => {
+            // Scan the occurrences in reverse source order and match the text of
+            // the last occurrence that is defined; a later occurrence overrides
+            // an earlier one only when it matched.
+            let mut matched = true;
+            for &group_idx in group_indices.iter().rev() {
+                let group = &s.groups[group_idx as usize];
+                if let Some(orig_range) = group.as_range() {
+                    if re.flags.icase {
+                        matched = matchers::backref_icase(input, dir, orig_range, &mut s.pos);
+                    } else {
+                        matched = matchers::backref(input, dir, orig_range, &mut s.pos);
+                    }
+                    break;
+                }
+            }
+            // No occurrence is defined, and therefore the match succeeds
+            // (ES6 21.2.2.9).
+            nextinsn_or_fail!(matched)
+        }
+
        &Insn::Lookahead {
            negate,
            start_group: _,
@@ -123,7 +123,7 @@ fn compute_start_predicate(n: &Node) -> Option<AbstractStartPredicate> {

        Node::Empty => arbitrary,
        Node::Goal => arbitrary,
-        Node::BackRef(..) => arbitrary,
+        Node::BackRef(..) | Node::BackRefLastDefined(..) => arbitrary,

        Node::CharSet(chars) => {
            // Pick the first bytes out.