feat: tarn show <file> N|A-B — bare line spec (sed Np reflex) (v0.8.1)

`tarn show file 5` / `file 5-9` now work as `--lines`, serving the
`sed -n '5p'` / `'5,9p'` reflex. Previously a second positional silently
overwrote the file path (`show file 5` tried to read a file named `5`); a
non-line-spec second positional now errors clearly. Same anti-silent-overwrite
shape as the v0.7.1 find fix; reuses parse_range.

168 tests (added show bare-line-spec regression), clippy + fmt clean.
Docs/whitepaper to v0.8.1; test count 167->168.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: OneMuppet

CHANGELOG.md

@@ -4,6 +4,11 @@ All notable changes to **tarn** are documented here. The format loosely follows
 [Keep a Changelog](https://keepachangelog.com/); this project is pre-1.0, so the
 surface may still shift.
 
+## [0.8.1]
+
+### `show` — accept a bare line spec (the sed `Np` reflex)
+- `tarn show <file> N` and `tarn show <file> A-B` now work as `--lines N` / `--lines A-B` — serving the `sed -n 'Np'` / `'A,Bp'` reflex directly. Previously a second positional silently overwrote the file path (so `show file 5` tried to read a file named `5`); a non-line-spec second positional now errors clearly instead.
+
 ## [0.8.0]
 
 ### Navigate — `locate`: find files by name

Cargo.lock

@@ -2,7 +2,7 @@
 # The crates.io package is `tarn-cli` (the name `tarn` is taken by an unrelated
 # crate); the installed command stays `tarn` via the [[bin]] below.
 name = "tarn-cli"
-version = "0.8.0"
+version = "0.8.1"
 edition = "2021"
 authors = ["David Borgenvik"]
 description = "A tiny, zero-dependency terminal editor and structural CLI toolkit for AI agents: navigate, search, edit, and patch code with exact line numbers and meaningful exit codes."

Cargo.toml

@@ -3,7 +3,7 @@
 ![tarn](assets/banner.svg)
 
 [![CI](https://github.com/OneMuppet/tarn/actions/workflows/ci.yml/badge.svg)](https://github.com/OneMuppet/tarn/actions/workflows/ci.yml)
- zero dependencies · 167 tests
+ zero dependencies · 168 tests
 
 </div>
 
@@ -131,7 +131,7 @@ returns structured data; on edits it returns a result object you can chain.
 
 ## Install
 
-These are live as of **v0.8.0**.
+These are live as of **v0.8.1**.
 
 **Homebrew:**
 
@@ -140,14 +140,14 @@ brew install onemuppet/tap/tarn
 ```
 
 **Prebuilt binary (no Rust toolchain needed)** — from the
-[v0.8.0 release](https://github.com/OneMuppet/tarn/releases/tag/v0.8.0):
+[v0.8.1 release](https://github.com/OneMuppet/tarn/releases/tag/v0.8.1):
 
 ```sh
 # macOS (Apple Silicon)
-curl -L https://github.com/OneMuppet/tarn/releases/download/v0.8.0/tarn-v0.8.0-aarch64-apple-darwin.tar.gz | tar xz
+curl -L https://github.com/OneMuppet/tarn/releases/download/v0.8.1/tarn-v0.8.1-aarch64-apple-darwin.tar.gz | tar xz
 
 # Linux (x86_64)
-curl -L https://github.com/OneMuppet/tarn/releases/download/v0.8.0/tarn-v0.8.0-x86_64-unknown-linux-gnu.tar.gz | tar xz
+curl -L https://github.com/OneMuppet/tarn/releases/download/v0.8.1/tarn-v0.8.1-x86_64-unknown-linux-gnu.tar.gz | tar xz
 ```
 
 Then put the extracted `tarn` on your `PATH`.
@@ -465,7 +465,7 @@ real multiplier for an agent is **repeated navigation + batched edits**:
 - The diff renderer trims the common prefix/suffix so a one-line change in a
   40k-line file diffs in ~26 ms instead of ~7 s.
 
-**Quality.** 27 commands, **167 tests**, gated by adversarial review on every
+**Quality.** 27 commands, **168 tests**, gated by adversarial review on every
 feature. The unsafe NEON path is **AddressSanitizer-clean** and the SIMD counter
 is **differential-tested** against a scalar oracle (900+ fuzz cases); its counts
 also match `rg`/`grep` on the benchmark corpus. Zero crate dependencies — std

README.md

@@ -417,7 +417,28 @@ fn cmd_show(args: &[String]) -> u8 {
             },
             "--head" => head = Some(next_usize_opt(args, &mut i).unwrap_or(20)),
             "--tail" => tail = Some(next_usize_opt(args, &mut i).unwrap_or(20)),
-            s if !s.starts_with("--") => file = Some(s),
+            s if !s.starts_with("--") => {
+                if file.is_none() {
+                    file = Some(s);
+                } else if lines.is_none() {
+                    // A bare line spec after the file (the sed `Np` / `A,Bp`
+                    // reflex): `show file 5` or `show file 5-9` ≡ `--lines`.
+                    // Don't silently overwrite the file with a second positional.
+                    match parse_range(s) {
+                        Some((a, b)) if a <= b => lines = Some((a, b)),
+                        _ => {
+                            eprintln!(
+                                "tarn: unexpected argument {s:?} — show takes <file> then an \
+                                 optional line spec (N or A-B), or --lines/--around/--head/--tail"
+                            );
+                            return EXIT_USAGE;
+                        }
+                    }
+                } else {
+                    eprintln!("tarn: unexpected extra argument {s:?}");
+                    return EXIT_USAGE;
+                }
+            }
             other => {
                 eprintln!("tarn: unknown flag {other}");
                 return EXIT_USAGE;
@@ -3557,6 +3578,23 @@ mod tests {
         let _ = std::fs::remove_file(&f);
     }
 
+    #[test]
+    fn show_accepts_bare_line_spec() {
+        // sed `Np`/`A,Bp` reflex: `show file N` / `file A-B` work as `--lines`,
+        // and a second positional no longer silently overwrites the file.
+        let f = std::env::temp_dir().join("tarn_test_show_linespec.txt");
+        std::fs::write(&f, "a\nb\nc\nd\n").unwrap();
+        let p = f.to_str().unwrap();
+        let s =
+            |v: &[&str]| -> u8 { cmd_show(&v.iter().map(|x| x.to_string()).collect::<Vec<_>>()) };
+        assert_eq!(s(&[p]), EXIT_OK); // file only
+        assert_eq!(s(&[p, "2"]), EXIT_OK); // bare line
+        assert_eq!(s(&[p, "2-3"]), EXIT_OK); // bare range
+        assert_eq!(s(&[p, "--lines", "1"]), EXIT_OK); // explicit form still works
+        assert_eq!(s(&[p, "notaspec"]), EXIT_USAGE); // garbage 2nd positional errors
+        let _ = std::fs::remove_file(&f);
+    }
+
     #[test]
     fn glob_to_regex_translates() {
         assert_eq!(glob_to_regex("*.env"), "^[^/]*\\.env$");

src/main.rs

@@ -126,7 +126,7 @@ <h1>The editor your agent<br>wishes it had<span class="dot">.</span></h1>
     <div class="stat"><div class="k">Dependencies</div><div class="v">0</div><div class="d">pure Rust std; mmap, threads & SIMD via core/libc — no crates</div></div>
     <div class="stat"><div class="k">Commands</div><div class="v">27</div><div class="d">navigate · read · edit · config · verify</div></div>
     <div class="stat"><div class="k">vs ripgrep</div><div class="v">1.3×</div><div class="d">faster on a single 380 MB file (~42 vs ~57 ms); parity on many files</div></div>
-    <div class="stat"><div class="k">Tests</div><div class="v">167</div><div class="d">+ adversarial review on every feature; ASan-clean</div></div>
+    <div class="stat"><div class="k">Tests</div><div class="v">168</div><div class="d">+ adversarial review on every feature; ASan-clean</div></div>
   </div>
 
   <div class="eyebrow copper" style="margin-top:40px">A field guide to a search-and-edit tool that earns an agent's trust</div>
@@ -201,7 +201,7 @@ <h3>Deterministic, guarded, chainable.</h3>
     previews; multi-file <code>apply</code>/<code>patch</code> are atomic with rollback; <code>--json</code> on read &amp; edit
     commands lets one command feed the next. An agent can act without a defensive re-read.</p>
   </div>
-  <div class="foot"><span>tarn · zero dependencies · 167 tests</span><span class="c">github.com/OneMuppet/tarn · 3 / 8</span></div>
+  <div class="foot"><span>tarn · zero dependencies · 168 tests</span><span class="c">github.com/OneMuppet/tarn · 3 / 8</span></div>
 </section>
 
 <!-- ============ 03 SPEED ============ -->
@@ -343,7 +343,7 @@ <h2>We stopped guessing. We counted.</h2>
   <div class="callout">
     <span class="e">What it isn't</span>
     <h3>A useful subset, named as one.</h3>
-    <p>The regex engine is a grep-ish subset, not PCRE — no backreferences, lookaround, or <code>\b</code>. Unsupported syntax errors <i>loudly</i> rather than matching the wrong thing, and literal substring stays the default. Structural <code>outline</code>/<code>defs</code> is heuristic, not a parser — multi-line signatures and keyword-less Java/C#/C/C++ methods are now handled by brace-balanced, string/comment-aware scanning; the residual known gap is C# <code>@"..."</code> verbatim strings. Built from the data, bounded by the truth — zero crates, 167 tests, every claim cashable.</p>
+    <p>The regex engine is a grep-ish subset, not PCRE — no backreferences, lookaround, or <code>\b</code>. Unsupported syntax errors <i>loudly</i> rather than matching the wrong thing, and literal substring stays the default. Structural <code>outline</code>/<code>defs</code> is heuristic, not a parser — multi-line signatures and keyword-less Java/C#/C/C++ methods are now handled by brace-balanced, string/comment-aware scanning; the residual known gap is C# <code>@"..."</code> verbatim strings. Built from the data, bounded by the truth — zero crates, 168 tests, every claim cashable.</p>
   </div>
   <div class="foot"><span>tarn · roadmap by measurement, not hunch</span><span class="c">github.com/OneMuppet/tarn · 7 / 8</span></div>
 </section>
@@ -366,7 +366,7 @@ <h2>Trust is the product.</h2>
     benchmark corpus. The unsafe NEON code is AddressSanitizer-clean.</p>
     <p><b>Edits never corrupt.</b> Line endings (LF/CRLF) and final-newline state are preserved on every
     edit; format-preserving config edits keep comments and key order; multi-file batches roll back as a unit.</p>
-    <p><b>167 tests, clippy &amp; fmt clean, CI green.</b> Every benchmark in this paper is reproducible from the
+    <p><b>168 tests, clippy &amp; fmt clean, CI green.</b> Every benchmark in this paper is reproducible from the
     repo. Where a number depended on the test machine, it's labelled. Nothing here is aspirational.</p>
   </div>