Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
//! GDB-driven trace capture for `anchor debugger --gdb`.
//!
//! Alternative to the register-tracing-file path. Each test thread's VM
//! blocks on a per-thread TCP port (sbpf's built-in gdb stub, activated
//! via the `debugger` feature on `solana-sbpf`). We drive the stub over
//! the GDB Remote Serial Protocol, single-stepping each invocation while
//! reading the full register set (including the pseudo-register at index
//! 12 that the sbpf target exposes as `InstructionCountRemaining` —
//! compute-unit remaining at each step).
//!
//! Files are written into the same `<profile_dir>/<test>/NNNN__txK.{regs,
//! insns,program_id,cu}` layout `TestNameCallback` produces, so the
//! existing arena + TUI code consumes them unchanged. The only new
//! artifact is `.cu` — 8 bytes per step, the VM's `cu_remaining` value
//! read via the gdb stub's register 12.
//!
//! ## CPI handling
//!
//! Each CPI frame constructs its own `EbpfVm`, which reads `VM_DEBUG_PORT`
//! and binds another listener on the same port (sbpf drops its listener
//! after `accept()`, so re-entrance is fine). The client side has to
//! notice that a step command on the outer connection is taking longer
//! than expected, open a second TCP connection to the same port, and
//! drive the inner session to completion — then the outer step reply
//! arrives, stepping continues. This mirrors the actual sbpf/agave call
//! stack: outer step → CPI syscall → inner VM exec → return → outer
//! step reply.

use {
    anyhow::{anyhow, Context, Result},
    std::{
        io::{BufRead, BufReader, Read, Write},
        net::TcpStream,
        os::unix::net::{UnixListener, UnixStream},
        path::{Path, PathBuf},
        sync::{
            atomic::{AtomicBool, Ordering},
            Arc, Mutex,
        },
        thread,
        time::{Duration, Instant},
    },
};

// sbpf gdb target layout. `g` dumps 12 x u64 = 96 bytes = 192 hex chars:
//   0..9   — Gpr (r0..r9)
//   10     — Sp
//   11     — Pc
// `InstructionCountRemaining` is a PSEUDO register at index 12; gdbstub
// doesn't include pseudo regs in `g` by convention — it requires a
// separate `p 0c` read. We issue that per step.
const REG_COUNT: usize = 12;
const REG_BYTES: usize = REG_COUNT * 8;
const REG_HEX_LEN: usize = REG_BYTES * 2;

/// Env var the gdb driver and the test process use to rendezvous on
/// the announce socket. Single source of truth — both the driver
/// (`std::env::set_var`) and `anchor-v2-testing` (which reads it) must
/// use this name.
pub const SOCKET_ENV: &str = "ANCHOR_GDB_SOCKET";

/// Owns the UDS listener at `<profile_dir>/gdb.sock` and the accept
/// thread that spawns one driver thread per VM announcement. Drop
/// signals stop, joins the accept thread, and removes the socket.
///
/// Decoupled from cargo invocation so both loose mode and Anchor.toml
/// mode can stand up the same listener+driver and just differ in how
/// the test process gets launched.
pub struct GdbDriver {
    profile_dir: PathBuf,
    sock_path: PathBuf,
    stop: Arc<AtomicBool>,
    accept: Option<thread::JoinHandle<Result<()>>>,
    captures: Arc<Mutex<Vec<PendingCapture>>>,
}

impl GdbDriver {
    pub fn sock_path(&self) -> &Path {
        &self.sock_path
    }

    fn shutdown(&mut self) {
        self.stop.store(true, Ordering::Release);
        if let Some(h) = self.accept.take() {
            let _ = h.join();
        }
        if let Ok(captures) = self.captures.lock() {
            if let Err(e) = write_pending_captures(&self.profile_dir, &captures) {
                eprintln!("gdb sidecar write error: {e}");
            }
        }
        let _ = std::fs::remove_file(&self.sock_path);
    }
}

impl Drop for GdbDriver {
    fn drop(&mut self) {
        self.shutdown();
    }
}

/// Bind the UDS listener and spawn the accept thread. Caller is
/// responsible for setting `ANCHOR_GDB_SOCKET` in the env of whatever
/// test process consumes the socket.
pub fn start_gdb_driver(profile_dir: &Path) -> Result<GdbDriver> {
    // Socket lives under the profile dir — same retention semantics as
    // the trace files themselves, cleaned on next debugger run.
    std::fs::create_dir_all(profile_dir).ok();
    let sock_path = profile_dir.join("gdb.sock");
    let _ = std::fs::remove_file(&sock_path); // clear stale
    let listener =
        UnixListener::bind(&sock_path).with_context(|| format!("bind {}", sock_path.display()))?;

    let stop = Arc::new(AtomicBool::new(false));
    let captures = Arc::new(Mutex::new(Vec::new()));

    // Accept thread: pull port announcements off the socket, spawn a
    // driver per announcement.
    let accept_stop = Arc::clone(&stop);
    let profile_dir_for_accept = profile_dir.to_path_buf();
    let captures_for_accept = Arc::clone(&captures);
    let accept_handle = thread::spawn(move || -> Result<()> {
        // Non-blocking accept loop so we can notice when the cargo test
        // process has exited and stop cleanly.
        listener
            .set_nonblocking(true)
            .context("set listener non-blocking")?;
        let mut drivers: Vec<thread::JoinHandle<()>> = Vec::new();
        while !accept_stop.load(Ordering::Acquire) {
            match listener.accept() {
                Ok((conn, _addr)) => {
                    let pd = profile_dir_for_accept.clone();
                    let captures = Arc::clone(&captures_for_accept);
                    drivers.push(thread::spawn(move || {
                        if let Err(e) = handle_announce(conn, &pd, captures) {
                            eprintln!("gdb driver error: {e}");
                        }
                    }));
                    // Reap finished drivers so the vec doesn't grow
                    // unbounded across long test runs.
                    drivers.retain(|h| !h.is_finished());
                }
                Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => {
                    thread::sleep(Duration::from_millis(20));
                }
                Err(e) => return Err(e).context("accept"),
            }
        }
        // Drain remaining drivers — outer-VM trace files must finish
        // writing before shutdown returns.
        for d in drivers {
            let _ = d.join();
        }
        Ok(())
    });

    Ok(GdbDriver {
        profile_dir: profile_dir.to_path_buf(),
        sock_path,
        stop,
        accept: Some(accept_handle),
        captures,
    })
}

/// Loose-mode entry point. Stands up the driver, then runs `cargo
/// test` with `ANCHOR_GDB_SOCKET` and `--test-threads=1`.
///
/// `--test-threads=1` is forced because `VM_DEBUG_PORT` is a
/// process-wide env var and sbpf reads it lazily inside each
/// `EbpfVm::new` — two test threads each running `svm()` with
/// different ports would have the last `set_var` win and both VMs
/// collide on bind. Proper per-thread ports would need sbpf to accept
/// the port through a non-env channel (fork).
#[allow(clippy::too_many_arguments)]
pub fn run_gdb_mode(
    cargo_cwd: &Path,
    current_package: Option<&str>,
    profile_feature: &str,
    profile_dir: &Path,
    test_filter: Option<&str>,
) -> Result<()> {
    let driver = start_gdb_driver(profile_dir)?;

    let mut cmd = std::process::Command::new("cargo");
    cmd.current_dir(cargo_cwd)
        .env(SOCKET_ENV, driver.sock_path())
        .env("ANCHOR_PROFILE_DIR", profile_dir)
        .env("RUST_TEST_THREADS", "1")
        .arg("test")
        .arg("--features")
        .arg(profile_feature);
    if let Some(pkg) = current_package {
        cmd.arg("-p").arg(pkg);
    }
    cmd.arg("--").arg("--test-threads=1");
    if let Some(filter) = test_filter {
        cmd.arg(filter);
    }
    let test_status = cmd.status().context("spawn cargo test")?;

    drop(driver);

    if !test_status.success() {
        return Err(anyhow!("cargo test failed"));
    }
    Ok(())
}

/// One announcement = one outer VM's port. Reads `"<port>\t<test_name>\n"`
/// off the UDS connection, connects to the port, drives the VM to
/// termination while probing for nested CPI listeners on the same port.
fn handle_announce(
    conn: UnixStream,
    profile_dir: &Path,
    captures: Arc<Mutex<Vec<PendingCapture>>>,
) -> Result<()> {
    let mut rdr = BufReader::new(conn);
    let mut line = String::new();
    rdr.read_line(&mut line).context("read announce")?;
    let line = line.trim_end();
    let (port_str, test_name) = line
        .split_once('\t')
        .ok_or_else(|| anyhow!("malformed announce: {line:?}"))?;
    let port: u16 = port_str.parse().context("parse port")?;

    let test_dir = profile_dir.join(sanitize(test_name));
    std::fs::create_dir_all(&test_dir).with_context(|| format!("create {}", test_dir.display()))?;

    // Retry-connect: sbpf may not have bound yet when we get the announce
    // (v2-testing announces before setting env + VM construct). A single
    // announced LiteSVM can execute multiple outer transactions, so keep
    // accepting sequential sessions on the same port until it goes quiet.
    let mut deadline = Duration::from_secs(5);
    while let Some(outer) = wait_for_connect(port, deadline) {
        let capture = drive_session(outer, port, test_name, 0)?;
        captures.lock().unwrap().push(PendingCapture {
            test_name: sanitize(test_name),
            capture,
        });
        deadline = Duration::from_millis(750);
    }
    Ok(())
}

struct PendingCapture {
    test_name: String,
    capture: CapturedInvocation,
}

struct CapturedInvocation {
    regs: Vec<u8>,
    insns: Vec<u8>,
    cu: Vec<u8>,
    children: Vec<CapturedInvocation>,
}

fn write_pending_captures(profile_dir: &Path, captures: &[PendingCapture]) -> Result<()> {
    let mut by_test = std::collections::BTreeMap::<&str, Vec<&CapturedInvocation>>::new();
    for pending in captures {
        by_test
            .entry(&pending.test_name)
            .or_default()
            .push(&pending.capture);
    }

    for (test_name, captures) in by_test {
        let test_dir = profile_dir.join(test_name);
        std::fs::create_dir_all(&test_dir)
            .with_context(|| format!("create {}", test_dir.display()))?;

        let mut flattened = Vec::new();
        for capture in captures {
            flatten_capture_postorder(capture, &mut flattened);
        }

        let mut regular = regular_invocation_stems(&test_dir)?;
        for capture in flattened {
            let step_count = capture.regs.len() / REG_BYTES;
            let regular_idx = regular
                .iter()
                .position(|r| !r.used && r.step_count == step_count)
                .or_else(|| regular.iter().position(|r| !r.used));
            let stem = if let Some(idx) = regular_idx {
                regular[idx].used = true;
                regular[idx].stem.clone()
            } else {
                let fallback = regular.len() + 1;
                test_dir.join(format!("{fallback:04}__tx1"))
            };
            write_capture_sidecars(capture, &stem)?;
        }
    }
    Ok(())
}

fn flatten_capture_postorder<'a>(
    capture: &'a CapturedInvocation,
    out: &mut Vec<&'a CapturedInvocation>,
) {
    for child in &capture.children {
        flatten_capture_postorder(child, out);
    }
    out.push(capture);
}

struct RegularInvocationStem {
    stem: PathBuf,
    step_count: usize,
    used: bool,
}

fn regular_invocation_stems(test_dir: &Path) -> Result<Vec<RegularInvocationStem>> {
    let mut found = Vec::new();
    if !test_dir.exists() {
        return Ok(Vec::new());
    }

    for entry in std::fs::read_dir(test_dir)? {
        let path = entry?.path();
        if path.extension().and_then(|e| e.to_str()) != Some("regs") {
            continue;
        }
        let Some(stem_str) = path.file_stem().and_then(|s| s.to_str()) else {
            continue;
        };
        if stem_str.ends_with(".gdb") {
            continue;
        }
        let Some((inv_part, tx_part)) = stem_str.split_once("__") else {
            continue;
        };
        let Some(tx_digits) = tx_part.strip_prefix("tx") else {
            continue;
        };
        let Ok(inv_seq) = inv_part.parse::<u32>() else {
            continue;
        };
        let Ok(tx_seq) = tx_digits.parse::<u32>() else {
            continue;
        };
        let step_count = std::fs::metadata(&path)
            .map(|m| m.len() as usize / REG_BYTES)
            .unwrap_or(0);
        found.push((
            tx_seq,
            inv_seq,
            RegularInvocationStem {
                stem: path.with_extension(""),
                step_count,
                used: false,
            },
        ));
    }

    found.sort_by_key(|(tx_seq, inv_seq, _)| (*tx_seq, *inv_seq));
    Ok(found.into_iter().map(|(_, _, stem)| stem).collect())
}

fn write_capture_sidecars(capture: &CapturedInvocation, stem: &Path) -> Result<()> {
    std::fs::write(stem.with_extension("gdb.regs"), &capture.regs)?;
    std::fs::write(stem.with_extension("gdb.insns"), &capture.insns)?;
    std::fs::write(stem.with_extension("gdb.cu"), &capture.cu)?;
    Ok(())
}

/// Drives one gdb session (one `EbpfVm::execute_program` invocation) to
/// termination. Recursive: if the session's step stalls (likely a CPI),
/// spawns a helper to connect to the same port and drive the nested
/// session, then resumes outer stepping.
fn drive_session(
    stream: TcpStream,
    port: u16,
    test_name: &str,
    depth: usize,
) -> Result<CapturedInvocation> {
    stream.set_nodelay(true).ok();
    let mut rsp = Rsp::new(stream);
    let started = Instant::now();
    let mut last_progress = started;

    rsp.send("QStartNoAckMode");
    let reply = rsp.recv();
    rsp.no_ack = reply == "OK";

    rsp.send("qSupported:multiprocess+;swbreak+;hwbreak+;vContSupported+");
    let _ = rsp.recv();

    rsp.send("?");
    let mut reply = rsp.recv();

    // Sidecar capture for the canonical invocation files produced by
    // TestNameCallback. We collect in memory so nested BPF CPIs can be
    // written in postorder, matching iterate_vm_traces' file stems.
    let mut capture = CapturedInvocation {
        regs: Vec::new(),
        insns: Vec::new(),
        cu: Vec::new(),
        children: Vec::new(),
    };

    let mut steps: u64 = 0;
    print_progress(test_name, port, depth, steps, started, false);
    loop {
        if reply.starts_with('W') || reply.starts_with('X') {
            break;
        }
        if !reply.starts_with('T') && !reply.starts_with('S') {
            eprintln!("unexpected stop reply: {reply}");
            break;
        }

        rsp.send("g");
        let regs_hex = rsp.recv();
        if regs_hex.is_empty() {
            break;
        }
        // Pseudo-register 12 = InstructionCountRemaining. Not included
        // in `g`'s dump, so read it separately.
        rsp.send("p0c");
        let cu_hex = rsp.recv();

        let (regs, pc, _cu_stub, insn) = decode_regs_hex(&regs_hex)?;
        let cu = decode_u64_hex(&cu_hex).unwrap_or(0);
        capture.regs.extend_from_slice(&regs);
        capture.cu.extend_from_slice(&cu.to_le_bytes());
        let insn_bytes = read_insn_at(&mut rsp, pc)?;
        capture.insns.extend_from_slice(&insn_bytes);
        let _ = insn;

        // Nested CPI probe: try connecting to the same port in a
        // background thread while we await the step reply. If the inner
        // VM has bound the port, our connect succeeds and we recurse.
        let nested_port = port;
        let nested_test_name = test_name.to_owned();
        let probe_cancel = Arc::new(AtomicBool::new(false));
        let probe_cancel_for_thread = Arc::clone(&probe_cancel);
        let probe = thread::spawn(move || -> Result<Option<CapturedInvocation>> {
            if let Some(inner) = probe_for_nested(
                nested_port,
                Duration::from_millis(250),
                &probe_cancel_for_thread,
            ) {
                return drive_session(inner, nested_port, &nested_test_name, depth + 1).map(Some);
            }
            Ok(None)
        });

        rsp.send("s");
        let step_reply = rsp.recv();
        probe_cancel.store(true, Ordering::Release);
        if let Ok(Ok(Some(child))) = probe.join() {
            capture.children.push(child);
        }
        if step_reply.is_empty() {
            // Stream closed — VM exited. Normal termination.
            break;
        }

        reply = step_reply;
        steps += 1;
        if last_progress.elapsed() >= Duration::from_secs(5) {
            print_progress(test_name, port, depth, steps, started, false);
            last_progress = Instant::now();
        }
    }
    print_progress(test_name, port, depth, steps, started, true);

    Ok(capture)
}

fn print_progress(
    test_name: &str,
    port: u16,
    depth: usize,
    steps: u64,
    started: Instant,
    done: bool,
) {
    let elapsed = started.elapsed().as_secs_f64().max(0.001);
    let rate = steps as f64 / elapsed;
    let indent = "  ".repeat(depth);
    let status = if done { "captured" } else { "stepping" };
    eprintln!(
        "{indent}gdb {status} {test_name} :{port} depth={depth} steps={steps} elapsed={:.1}s \
         rate={:.1}/s",
        elapsed, rate,
    );
}

fn wait_for_connect(port: u16, deadline: Duration) -> Option<TcpStream> {
    let start = Instant::now();
    while start.elapsed() < deadline {
        if let Ok(s) = TcpStream::connect_timeout(
            &format!("127.0.0.1:{port}").parse().ok()?,
            Duration::from_millis(50),
        ) {
            return Some(s);
        }
        thread::sleep(Duration::from_millis(5));
    }
    None
}

fn probe_for_nested(port: u16, window: Duration, cancel: &AtomicBool) -> Option<TcpStream> {
    let start = Instant::now();
    while start.elapsed() < window && !cancel.load(Ordering::Acquire) {
        if let Ok(s) = TcpStream::connect(("127.0.0.1", port)) {
            return Some(s);
        }
        thread::yield_now();
    }
    None
}

fn decode_regs_hex(hex: &str) -> Result<([u8; REG_BYTES], u64, u64, [u8; 8])> {
    if hex.len() < REG_HEX_LEN {
        return Err(anyhow!(
            "register hex too short: {} bytes, payload={hex:?}",
            hex.len()
        ));
    }
    let mut regs = [0u8; REG_BYTES];
    for (i, pair) in hex.as_bytes()[..REG_HEX_LEN].chunks_exact(2).enumerate() {
        let s = std::str::from_utf8(pair).unwrap();
        regs[i] = u8::from_str_radix(s, 16).context("invalid hex")?;
    }
    // r11 = pc in sbpf's gdb target (last u64 in `g`'s dump).
    let pc = u64::from_le_bytes(regs[11 * 8..12 * 8].try_into().unwrap());
    let cu = 0; // CU read separately via `p 0c`.
    let insn = [0u8; 8]; // real bytes come from ELF, not gdb
    Ok((regs, pc, cu, insn))
}

fn decode_u64_hex(hex: &str) -> Option<u64> {
    if hex.len() < 16 {
        return None;
    }
    let mut bytes = [0u8; 8];
    for (i, pair) in hex.as_bytes()[..16].chunks_exact(2).enumerate() {
        bytes[i] = u8::from_str_radix(std::str::from_utf8(pair).ok()?, 16).ok()?;
    }
    Some(u64::from_le_bytes(bytes))
}

fn read_insn_at(rsp: &mut Rsp, pc: u64) -> Result<[u8; 8]> {
    // The sbpf gdb stub reports PC as the program virtual byte address.
    // Reading memory at that same address returns the raw instruction bytes.
    rsp.send(&format!("m{pc:x},8"));
    decode_bytes_hex::<8>(&rsp.recv())
}

fn decode_bytes_hex<const N: usize>(hex: &str) -> Result<[u8; N]> {
    if hex.len() < N * 2 {
        return Err(anyhow!(
            "memory hex too short: {} bytes, payload={hex:?}",
            hex.len()
        ));
    }
    let mut out = [0u8; N];
    for (i, pair) in hex.as_bytes()[..N * 2].chunks_exact(2).enumerate() {
        out[i] =
            u8::from_str_radix(std::str::from_utf8(pair).unwrap(), 16).context("invalid hex")?;
    }
    Ok(out)
}

fn sanitize(s: &str) -> String {
    s.chars()
        .map(|c| {
            if c.is_ascii_alphanumeric() || c == '_' {
                c
            } else {
                '_'
            }
        })
        .collect()
}

// ---------------------------------------------------------------------------
// Minimal GDB Remote Serial Protocol client. Just enough for our step loop.

struct Rsp {
    stream: TcpStream,
    buf: Vec<u8>,
    no_ack: bool,
}

impl Rsp {
    fn new(stream: TcpStream) -> Self {
        Self {
            stream,
            buf: Vec::with_capacity(512),
            no_ack: false,
        }
    }

    fn send(&mut self, payload: &str) {
        let mut cksum: u32 = 0;
        for b in payload.bytes() {
            cksum = cksum.wrapping_add(b as u32);
        }
        let pkt = format!("${payload}#{:02x}", cksum & 0xff);
        let _ = self.stream.write_all(pkt.as_bytes());
        let _ = self.stream.flush();
        if !self.no_ack {
            let mut one = [0u8; 1];
            let _ = self.stream.read_exact(&mut one);
        }
    }

    fn recv(&mut self) -> String {
        let mut one = [0u8; 1];
        loop {
            if self.stream.read_exact(&mut one).is_err() {
                return String::new();
            }
            if one[0] == b'$' {
                break;
            }
        }
        self.buf.clear();
        loop {
            if self.stream.read_exact(&mut one).is_err() {
                return String::new();
            }
            if one[0] == b'#' {
                break;
            }
            self.buf.push(one[0]);
        }
        let mut cksum = [0u8; 2];
        let _ = self.stream.read_exact(&mut cksum);
        if !self.no_ack {
            let _ = self.stream.write_all(b"+");
            let _ = self.stream.flush();
        }
        // Decode GDB RLE: a `*` followed by a count byte means "repeat the
        // previous char N more times", where N = count_byte - 28. This is
        // what sbpf's stub uses to shrink register dumps full of zeros
        // down from 208 bytes to ~45.
        let decoded = rle_decode(&self.buf);
        String::from_utf8_lossy(&decoded).into_owned()
    }
}

fn rle_decode(src: &[u8]) -> Vec<u8> {
    // GDB RSP RLE: `X*N` where N is an ASCII char, means the total
    // run length of X (including the literal X) is `N - 29 + 1`. So
    // we push `N - 29` additional copies of the preceding char.
    let mut out = Vec::with_capacity(src.len() * 4);
    let mut i = 0;
    while i < src.len() {
        let c = src[i];
        if c == b'*' && i + 1 < src.len() && !out.is_empty() {
            let n = (src[i + 1] as usize).saturating_sub(29);
            let last = *out.last().unwrap();
            for _ in 0..n {
                out.push(last);
            }
            i += 2;
        } else {
            out.push(c);
            i += 1;
        }
    }
    out
}

#[cfg(test)]
mod tests {
    use {super::*, std::fmt::Write as _, tempfile::tempdir};

    fn bytes_to_hex(bytes: &[u8]) -> String {
        let mut out = String::with_capacity(bytes.len() * 2);
        for byte in bytes {
            write!(&mut out, "{byte:02x}").unwrap();
        }
        out
    }

    fn regs_hex(regs: [u64; REG_COUNT]) -> String {
        let bytes = regs
            .iter()
            .flat_map(|reg| reg.to_le_bytes())
            .collect::<Vec<_>>();
        bytes_to_hex(&bytes)
    }

    #[test]
    fn decode_regs_hex_reads_pc_from_r11_little_endian() {
        let mut regs = [0u64; REG_COUNT];
        regs[0] = 0x11;
        regs[11] = 0x1122_3344_5566_7788;

        let (raw, pc, cu, insn) = decode_regs_hex(&regs_hex(regs)).unwrap();

        assert_eq!(pc, 0x1122_3344_5566_7788);
        assert_eq!(cu, 0);
        assert_eq!(insn, [0; 8]);
        assert_eq!(u64::from_le_bytes(raw[0..8].try_into().unwrap()), 0x11);
    }

    #[test]
    fn decode_bytes_hex_rejects_short_or_invalid_payload() {
        assert!(decode_bytes_hex::<8>("00").is_err());
        assert!(decode_bytes_hex::<1>("zz").is_err());
    }

    #[test]
    fn rle_decode_expands_gdb_rsp_runs() {
        assert_eq!(rle_decode(b"A* B"), b"AAAAB".to_vec());
    }

    #[test]
    fn regular_invocation_stems_ignores_gdb_sidecars_and_sorts_by_tx_inv() {
        let dir = tempdir().unwrap();
        std::fs::write(dir.path().join("0002__tx1.regs"), vec![0u8; REG_BYTES]).unwrap();
        std::fs::write(dir.path().join("0001__tx1.regs"), vec![0u8; REG_BYTES * 2]).unwrap();
        std::fs::write(dir.path().join("0001__tx2.regs"), vec![0u8; REG_BYTES]).unwrap();
        std::fs::write(dir.path().join("0001__tx1.gdb.regs"), vec![0u8; REG_BYTES]).unwrap();

        let stems = regular_invocation_stems(dir.path()).unwrap();

        assert_eq!(stems.len(), 3);
        assert_eq!(stems[0].stem.file_name().unwrap(), "0001__tx1");
        assert_eq!(stems[0].step_count, 2);
        assert_eq!(stems[1].stem.file_name().unwrap(), "0002__tx1");
        assert_eq!(stems[2].stem.file_name().unwrap(), "0001__tx2");
    }
}