sed: point the invalid back-reference caret at the reference

src/sed/command.rs

@@ -137,6 +137,8 @@ pub enum RegexMode {
 pub struct ReplacementTemplate {
     pub parts: Vec<ReplacementPart>,
     pub max_group_number: usize, // Highest used group number (e.g. 8 for \8)
+    /// Where the first reference to `max_group_number` appears in the script.
+    pub max_group_location: Option<ScriptLocation>,
 }
 
 impl Default for ReplacementTemplate {
@@ -161,6 +163,7 @@ impl ReplacementTemplate {
         Self {
             parts,
             max_group_number: max_group_number.try_into().unwrap(),
+            max_group_location: None,
         }
     }
 

src/sed/compiler.rs

@@ -752,6 +752,7 @@ pub fn compile_replacement(
 ) -> UResult<ReplacementTemplate> {
     let mut parts = Vec::new();
     let mut literal = Vec::new();
+    let mut max_group: Option<(u32, ScriptLocation)> = None;
 
     let delimiter = line.current();
     line.advance();
@@ -787,6 +788,12 @@ pub fn compile_replacement(
                             if ref_num == 0 {
                                 parts.push(ReplacementPart::WholeMatch);
                             } else {
+                                if max_group.as_ref().is_none_or(|(n, _)| ref_num > *n) {
+                                    // Point at the backslash, one byte back.
+                                    let mut location = ScriptLocation::at_position(lines, line);
+                                    location.column_number -= 1;
+                                    max_group = Some((ref_num, location));
+                                }
                                 parts.push(ReplacementPart::Group(ref_num));
                             }
                             line.advance();
@@ -837,7 +844,9 @@ pub fn compile_replacement(
                     if !literal.is_empty() {
                         parts.push(ReplacementPart::Literal(literal));
                     }
-                    return Ok(ReplacementTemplate::new(parts));
+                    let mut template = ReplacementTemplate::new(parts);
+                    template.max_group_location = max_group.map(|(_, location)| location);
+                    return Ok(template);
                 }
 
                 _ => {
@@ -915,12 +924,14 @@ fn compile_subst_command(
     .map(Rc::new);
 
     // Catch invalid group references at compile time, if possible.
+    // `compile_replacement` records a location for every `\N` it reads,
+    // so the location is only absent when there is no group reference.
     if let Some(regex) = &subst.regex
+        && let Some(location) = &subst.replacement.max_group_location
         && subst.replacement.max_group_number > regex.captures_len() - 1
     {
-        return compilation_error(
-            lines,
-            line,
+        return semantic_error(
+            location,
             format!(
                 "invalid reference \\{} on `s' command's RHS",
                 subst.replacement.max_group_number

tests/by-util/test_sed.rs

@@ -1649,6 +1649,45 @@ fn diagnostics_reach_errors_raised_after_the_line_was_read() {
         ));
 }
 
+#[test]
+fn diagnostics_underline_the_invalid_back_reference() {
+    // The message is raised after the flags are read; the caret still
+    // points at the first `\3`, not past the end of the command.
+    new_ucmd!()
+        .env("UUTILS_DIAG", "always")
+        .args(&[r"s/\(a\)/x\3\1\3/"])
+        .fails()
+        .code_is(1)
+        .stderr_is(concat!(
+            "sed: <script argument 1>:1:10: error: invalid reference \\3 on `s' command's RHS\n",
+            "   ╭─[ <script argument 1>:1:10 ]\n",
+            "   │\n",
+            r" 1 │ s/\(a\)/x\3\1\3/",
+            "\n",
+            "   │          ─\n",
+            "───╯\n",
+        ));
+}
+
+#[test]
+fn diagnostics_underline_the_invalid_back_reference_on_a_later_line() {
+    // The replacement continues on line 2: the caret follows it there.
+    new_ucmd!()
+        .env("UUTILS_DIAG", "always")
+        .args(&["-e", "s/\\(a\\)/x\\\ny\\2/"])
+        .fails()
+        .code_is(1)
+        .stderr_is(concat!(
+            "sed: <script argument 1>:2:2: error: invalid reference \\2 on `s' command's RHS\n",
+            "   ╭─[ <script argument 1>:2:2 ]\n",
+            "   │\n",
+            r" 2 │ y\2/",
+            "\n",
+            "   │  ─\n",
+            "───╯\n",
+        ));
+}
+
 #[test]
 fn diagnostics_line_up_under_wide_characters() {
     // `日本` takes four columns: the underline must sit under `q`.