logger: indent the caret relative to the windowed line excerpt

src/ast/lib.rs

@@ -651,6 +651,11 @@ pub struct Location {
     pub length: usize,
     // TODO: document or remove
     pub offset: usize,
+    /// 0-based column (UTF-16 units, like `column`) at which `line_text`
+    /// starts in the source line. Non-zero when `line_text` is a window of a
+    /// longer line; `write_format` subtracts it from `column` to place the
+    /// caret under the excerpt.
+    pub line_text_start_column: usize,
 
     /// 1-based line number.
     /// Line <= 0 means there is no line and column information.
@@ -679,6 +684,7 @@ impl Clone for Location {
             length: self.length,
             line_text: self.line_text.as_deref().map(|t| Cow::Owned(t.to_vec())),
             offset: self.offset,
+            line_text_start_column: self.line_text_start_column,
         }
     }
 }
@@ -691,6 +697,7 @@ impl Default for Location {
             line_text: None,
             length: 0,
             offset: 0,
+            line_text_start_column: 0,
             line: 0,
             column: 0,
         }
@@ -738,6 +745,7 @@ impl Location {
             length: self.length,
             line_text: self.line_text.as_deref().map(|t| Cow::Owned(t.to_vec())),
             offset: self.offset,
+            line_text_start_column: self.line_text_start_column,
         }
     }
 
@@ -759,6 +767,7 @@ impl Location {
             length: length as usize,
             line_text: line_text.map(Cow::Borrowed),
             offset: length as usize,
+            line_text_start_column: 0,
         }
     }
 
@@ -792,6 +801,7 @@ impl Location {
                     length: 0,
                     line_text: Some(Cow::Borrowed(b"")),
                     offset: 0,
+                    line_text_start_column: 0,
                 });
             }
             let data = match tracker {
@@ -800,11 +810,15 @@ impl Location {
             };
             let mut full_line = &source.contents[data.line_start..data.line_end];
             // Window a long line to ~120 bytes around the error. Bounds are
-            // BYTE offsets; the gate keeps the original shape (no left trim for
-            // an error in the last 80 bytes) so `write_format`'s caret aligns.
+            // BYTE offsets. `line_text_start_column` records how many columns
+            // the window drops on the left so `write_format`'s caret aligns.
+            // Consumers that only know `column` (the bake overlay,
+            // `BuildMessage.position`) rely on the gate not left-trimming an
+            // error in the last 80 bytes of a line.
             let offset_in_line = clamp_error_offset(&source.contents, r.loc)
                 .saturating_sub(data.line_start)
                 .min(full_line.len());
+            let mut line_text_start_column = 0;
             if full_line.len() > 80 + offset_in_line {
                 let mut lo = offset_in_line.saturating_sub(40);
                 let mut hi = (offset_in_line + 80).min(full_line.len());
@@ -816,6 +830,13 @@ impl Location {
                 {
                     hi += 1;
                 }
+                if lo > 0 {
+                    // Count the <= 40 kept bytes with the scanner that produced
+                    // `column_count`, instead of rescanning the dropped prefix.
+                    let mut kept = ErrorPositionState::default();
+                    kept.advance(full_line, lo, offset_in_line);
+                    line_text_start_column = data.column_count.saturating_sub(kept.column_number);
+                }
                 full_line = &full_line[lo..hi];
             }
 
@@ -832,11 +853,11 @@ impl Location {
                 // `source_backing` in `Transpiler::parse_*` is RAII and
                 // drops on the parse-error path *before* `process_fetch_log`
                 // clones the `Msg` into a `BuildMessage`, so own the bytes here
-                // instead of borrowing `source.contents`. `full_line` is
-                // bounded (≤ ~120 bytes) and only materialized on diagnostic
-                // paths.
-                line_text: Some(Cow::Owned(bun_core::trim_left(full_line, b"\n\r").to_vec())),
+                // instead of borrowing `source.contents`. Only materialized on
+                // diagnostic paths.
+                line_text: Some(Cow::Owned(full_line.to_vec())),
                 offset: usize::try_from(r.loc.start.max(0)).expect("int cast"),
+                line_text_start_column,
             });
         }
         None
@@ -981,8 +1002,13 @@ impl Data {
                 let line_text_right_trimmed = bun_core::trim_right(line_text_, b" \r\n\t");
                 let line_text = bun_core::trim_left(line_text_right_trimmed, b"\n\r");
                 if location.column > 0 && !line_text.is_empty() {
+                    // The caret sits under the printed excerpt, so its indent
+                    // is bounded by the excerpt, never by the full column.
                     let mut line_offset_for_second_line: usize =
-                        usize::try_from(location.column - 1).expect("int cast");
+                        usize::try_from(location.column - 1)
+                            .expect("int cast")
+                            .saturating_sub(location.line_text_start_column)
+                            .min(line_text.len());
 
                     if location.line > -1 {
                         let bold = matches!(kind, Kind::Err | Kind::Warn);
@@ -2454,11 +2480,7 @@ impl ErrorPositionState {
 
     fn to_error_position(self, line_end: usize) -> ErrorPosition {
         ErrorPosition {
-            line_start: if self.line_start > 0 {
-                self.line_start - 1
-            } else {
-                self.line_start
-            },
+            line_start: self.line_start,
             line_end,
             line_count: self.line_count,
             column_count: self.column_number,

test/cli/install/bun-lock.test.ts

@@ -1064,6 +1064,47 @@ it("prints an actionable error for a lockfile version newer than this build supp
   expect(await exited).toBe(0);
 });
 
+it("indents the caret under the printed excerpt when a one-line bun.lock has many warnings", async () => {
+  const count = 300;
+  const lock =
+    `{\n  "lockfileVersion": 1,\n  "workspaces": { "": { "name": "one-line-lockfile" } },\n  "packages": { ` +
+    Array.from({ length: count }, (_, i) => `"p${i}": ["p${i}@1.0.0", "", {}, "sha512-x"]`).join(", ") +
+    ` }\n}\n`;
+  using dir = tempDir("one-line-lockfile", {
+    "package.json": JSON.stringify({ name: "one-line-lockfile", dependencies: {} }),
+    "bun.lock": lock,
+  });
+
+  await using proc = spawn({
+    cmd: [bunExe(), "install"],
+    cwd: String(dir),
+    env,
+    stdout: "pipe",
+    stderr: "pipe",
+  });
+  const [, err, exitCode] = await Promise.all([proc.stdout.text(), proc.stderr.text(), proc.exited]);
+
+  const lines = err.split("\n");
+  const carets = lines.filter(line => /^ *\^$/.test(line));
+  expect(carets.length).toBe(count);
+
+  // Each caret points at the `"` that opens the bad integrity string in the
+  // excerpt above it. Before the fix it was padded to the column in the full
+  // line, so the caret lines alone grew quadratically with the entry count.
+  for (let i = 0; i < lines.length; i++) {
+    if (!/^ *\^$/.test(lines[i])) continue;
+    const excerpt = lines[i - 1];
+    expect(excerpt).toMatch(/^4 \| /);
+    const caretAt = lines[i].indexOf("^");
+    expect(caretAt).toBeLessThan(excerpt.length);
+    expect(excerpt.slice(caretAt, caretAt + 8)).toBe('"sha512-');
+  }
+  // Linear in the entry count: each warning prints a ~120-byte excerpt, except
+  // the ones in the last 80 bytes of the line, which print the whole line.
+  expect(err.length).toBeLessThan(count * 500 + 4 * lock.length);
+  expect(exitCode).toBe(0);
+});
+
 async function installWithHandEditedOverrides(overrides: Record<string, unknown>) {
   const { packageDir, packageJson } = await registry.createTestDir();
   const lockfile = JSON.stringify(

test/js/bun/transpiler/parse-error-column.test.ts

@@ -141,22 +141,111 @@ test.concurrent("long non-ASCII line's lineText window does not split a UTF-8 se
   });
 });
 
-test.concurrent("CLI caret stays under the token for an error at the end of a long line", async () => {
-  // `write_format` offsets the caret by `column - 1` with no knowledge of any
-  // left-trim, so the window gate must not left-trim this case.
-  using dir = tempDir("parse-col-caret", {
-    "long.js": Buffer.alloc(150, "a").toString() + "]",
-  });
+const fill = (count: number, char: string) => Buffer.alloc(count, char).toString();
+
+/**
+ * Runs `bun <args>` in a directory holding `files` and returns every source
+ * excerpt the logger printed (a `N | text` line followed by a caret line) along
+ * with the column the `^` landed on. The excerpt may be a window of a long line;
+ * the caret must land on the offending token inside the printed excerpt.
+ */
+async function printedExcerpts(args: string[], files: Record<string, string>) {
+  using dir = tempDir("parse-col-caret", files);
   await using proc = Bun.spawn({
-    cmd: [bunExe(), "build", "long.js"],
-    env: { ...bunEnv, NO_COLOR: "1" },
+    cmd: [bunExe(), ...args],
+    env: bunEnv,
     cwd: String(dir),
     stdout: "pipe",
     stderr: "pipe",
   });
-  const [, stderr] = await Promise.all([proc.stdout.text(), proc.stderr.text(), proc.exited]);
-  const lines = stderr.split("\n");
-  const textLine = lines.find(l => l.includes("]"))!;
-  const caretLine = lines.find(l => l.trimEnd().endsWith("^"))!;
-  expect({ token: textLine.indexOf("]"), caret: caretLine.indexOf("^") }).toEqual({ token: 154, caret: 154 });
+  const [stderr] = await Promise.all([proc.stderr.text(), proc.stdout.text(), proc.exited]);
+  const lines = stderr.split(/\r?\n/);
+  const excerpts: { excerpt: string; caret: number }[] = [];
+  for (let i = 0; i + 1 < lines.length; i++) {
+    if (/^\d+ \| /.test(lines[i]) && /^ *\^$/.test(lines[i + 1])) {
+      excerpts.push({ excerpt: lines[i], caret: lines[i + 1].indexOf("^") });
+    }
+  }
+  return { stderr, excerpts };
+}
+
+test.concurrent("CLI caret stays under the token for an error at the end of a long line", async () => {
+  // An error in the last 80 bytes of a line is never left-trimmed.
+  const { excerpts } = await printedExcerpts(["build", "long.js"], { "long.js": fill(150, "a") + "]" });
+  expect(excerpts).toEqual([{ excerpt: "1 | " + fill(150, "a") + "]", caret: 4 + 150 }]);
+});
+
+test.concurrent.each([["build"], ["run"]])(
+  "bun %s: CLI caret stays under the token when the excerpt is left-trimmed",
+  async subcommand => {
+    // `]` is at byte 100 of a 201-byte line, so the excerpt is the 120 bytes
+    // around it (40 before, 80 after) and the caret belongs 40 characters in,
+    // not at the token's column in the full line (which is past the end of
+    // the excerpt).
+    const { excerpts } = await printedExcerpts([subcommand, "long.js"], {
+      "long.js": "let ok = 1;\n" + fill(100, "a") + "]" + fill(100, "b") + "\n",
+    });
+    expect(excerpts).toEqual([{ excerpt: "2 | " + fill(40, "a") + "]" + fill(79, "b"), caret: 4 + 40 }]);
+  },
+);
+
+test.concurrent("CLI caret counts the left-trimmed prefix in columns, not bytes", async () => {
+  // U+00E9 is 2 UTF-8 bytes but 1 column (Buffer.alloc fills by bytes, so
+  // fill(200) is 100 characters). `]` is at byte 200 / column 101 of a
+  // 401-byte line; the window keeps the 40 bytes (20 characters) before it,
+  // so the caret belongs 20 characters in.
+  const { excerpts } = await printedExcerpts(["build", "long.js"], {
+    "long.js": fill(200, "\u00E9") + "]" + fill(200, "\u00E9"),
+  });
+  expect(excerpts).toEqual([{ excerpt: "1 | " + fill(40, "\u00E9") + "]" + fill(80, "\u00E9"), caret: 4 + 20 }]);
+});
+
+test.concurrent("CLI caret counts a left-trimmed astral prefix in UTF-16 units, like the column", async () => {
+  // U+10400 (a valid identifier character) is 4 UTF-8 bytes and 2 UTF-16
+  // units, and columns count UTF-16 units. 30 of them put `]` at byte 120 /
+  // column 61; the window keeps the 10 characters (20 units) before it, so
+  // the caret belongs 20 units in (10 if the trimmed width were counted in
+  // characters, 60 if it were not subtracted at all).
+  const { excerpts } = await printedExcerpts(["build", "long.js"], {
+    "long.js": fill(120, "\u{10400}") + "]" + fill(100, "b"),
+  });
+  expect(excerpts).toEqual([{ excerpt: "1 | " + fill(40, "\u{10400}") + "]" + fill(79, "b"), caret: 4 + 20 }]);
+});
+
+test.concurrent("CLI caret is aligned for both the error and its note on a long line", async () => {
+  // The redeclaration at byte 176 is left-trimmed; the note pointing at the
+  // original declaration (byte 6) is windowed without a left trim.
+  const source = "const x = 1; /* " + fill(150, "p") + " */ const x = 2; " + fill(100, "q") + "\n";
+  const { excerpts } = await printedExcerpts(["build", "long.js"], { "long.js": source });
+  expect(excerpts).toEqual([
+    { excerpt: "1 | " + fill(30, "p") + " */ const x = 2; " + fill(73, "q"), caret: 4 + 30 + " */ const ".length },
+    { excerpt: "1 | const x = 1; /* " + fill(70, "p"), caret: 4 + "const ".length },
+  ]);
+});
+
+test.concurrent("bun install points the caret at the token inside a long package.json line", async () => {
+  // The `}` is at column 179 of a 379-byte line. The excerpt starts 40 bytes
+  // before it (the caret used to be indented by the full 178 columns), while
+  // the `at file:line:col` suffix still reports the real column.
+  const { stderr, excerpts } = await printedExcerpts(["install"], {
+    "package.json": '{"name":"x",' + fill(150, " ") + '"dependencies": }' + fill(200, " ") + "\n",
+  });
+  expect(excerpts).toEqual([
+    { excerpt: "1 | " + fill(24, " ") + '"dependencies": }', caret: 4 + 24 + '"dependencies": '.length },
+  ]);
+  expect(stderr).toContain("package.json:1:179");
 });
+
+test.concurrent(
+  "excerpt of a line following a U+2028 line separator starts at the line's first character",
+  async () => {
+    // The logger used to slice the line from one byte before its start to pick
+    // up (and later trim) a preceding `\n`; after a three-byte U+2028 that byte
+    // is a stray UTF-8 continuation byte, which was printed before the text and
+    // pushed it one cell to the right of the caret.
+    const { excerpts } = await printedExcerpts(["build", "ls.js"], {
+      "ls.js": "let a = 1;\u2028let b = 2; ]",
+    });
+    expect(excerpts).toEqual([{ excerpt: "2 | let b = 2; ]", caret: 4 + "let b = 2; ".length }]);
+  },
+);