# Plan: Full `latprobe` Python Package

**Timestamp:** 2026-07-01 12:23
**Status:** Complete

## Goal

Build `python/latprobe/` — a packaged Python port of the Go `latprobe` CLI that
mirrors its behavior: flags, sampling, cross-URL concurrency, JSON output, and
distinct exit codes.

## Package layout

```
python/
  latprobe/
    __init__.py     — package marker
    __main__.py     — sys.exit(cli.run(sys.argv[1:], sys.stdout, sys.stderr))
    probe.py        — measure(url, opts) -> Result  (raw socket timing)
    aggregate.py    — summarize(results) -> Aggregate  (min/avg/max per phase)
    cli.py          — run(args, stdout, stderr) -> int  (argparse + rendering)
    duration.py     — parse_duration("10s") -> float seconds
  tests/
    test_probe.py   — unittest: success, DNS fail, connect refused, timeout, bad scheme
    test_cli.py     — unittest: drives cli.run() in-process, all branches + JSON
```

## Key design decisions

- **`probe.py`**: reuses the raw-socket technique from `phases.py` with an
  `Options(timeout)` parameter. Partial phases preserved on failure.
- **`aggregate.py`**: `summarize(List[Result]) -> Aggregate` computes
  `PhaseStats(min_ms, avg_ms, max_ms, present)` per phase. Uses only `succeeded`
  results.
- **`duration.py`**: parses `"10s"`, `"500ms"`, `"2m"`, bare numbers → float seconds.
- **`cli.py`**: injectable `run(args, stdout, stderr) -> int` seam for testability.
  - `_Parser` subclasses `argparse.ArgumentParser` with custom `print_help`,
    `print_usage`, `error`, `exit` to redirect all output to injected streams
    and raise `_ArgExit` instead of calling `sys.exit`.
  - `ThreadPoolExecutor` for cross-URL concurrency; `executor.map` preserves
    input order.
  - Four text-rendering branches: single-success, multi-sample aggregate,
    all-failed, mixed.
  - Failure summary groups errors by `(phase, message)` preserving insertion
    order; `N ×` prefix when grouped.

## CLI flags

| Flag | Default | Description |
|------|---------|-------------|
| `urls` (positional, `+`) | required | One or more URLs |
| `-n`/`--count` | 1 | Requests per URL |
| `-c`/`--concurrency` | 0 (auto) | Max parallel URLs; auto = `min(numURLs, 8)` |
| `--timeout` | `10s` | Per-request timeout (parsed via `duration.py`) |
| `--fail` | off | Exit non-zero on HTTP status >= 400 |
| `--json` | off | Output as JSON array instead of text |

## Exit codes

| Code | Meaning |
|------|---------|
| 0 | All probes succeeded (with `--fail`: no 4xx/5xx) |
| 1 | Usage / argument error |
| 2 | DNS failure |
| 3 | TCP connect failure |
| 4 | Timeout |
| 5 | TLS error |
| 6 | HTTP status >= 400 (`--fail` only) |

Worst (highest) code across all URLs wins.

## JSON schema

```json
[
  {
    "url": "https://example.com",
    "status": 200,
    "succeeded": 3,
    "failed": 2,
    "phases": {
      "dns":      {"min_ms": 1.23, "avg_ms": 1.45, "max_ms": 1.67},
      "connect":  {"min_ms": ...},
      "tls":      {"min_ms": ...},
      "ttfb":     {"min_ms": ...},
      "transfer": {"min_ms": ...},
      "total":    {"min_ms": ...}
    },
    "errors": [
      {"phase": "connect", "count": 2, "message": "..."}
    ]
  }
]
```

- `phases` omitted when `succeeded == 0`
- Per-phase key omitted when no successful sample had that phase (e.g. `tls` for `http://`)
- `errors` omitted when `failed == 0`

## Text output format

Single sample:
```
https://example.com  (200)
  DNS lookup     :     18.87 ms
  TCP connect    :      9.76 ms
  TLS handshake  :     14.71 ms
  Server (TTFB)  :     67.07 ms
  Transfer       :      0.14 ms
  ─────────────────────────────
  Total          :    118.39 ms
```

Aggregate (-n 3):
```
https://example.com  (200, 3 samples)
                          min        avg        max
  DNS lookup     :    2.36 ms    3.51 ms    5.00 ms
  TCP connect    :   10.25 ms   17.45 ms   31.61 ms
  TLS handshake  :   14.04 ms   41.60 ms   96.50 ms
  Server (TTFB)  :   63.30 ms   66.72 ms   69.78 ms
  Transfer       :    0.26 ms    0.38 ms    0.51 ms
  ─────────────────────────────────────────────────
  Total          :  101.33 ms  139.19 ms  211.99 ms
```

## Testing approach

- `test_probe.py`: real `http.server.HTTPServer` in a daemon thread; closed port
  for connection-refused; `.invalid` TLD for DNS failure; black-hole socket (accepts
  TCP, never sends data) to trigger TTFB timeout.
- `test_cli.py`: in-process `cli.run(args, io.StringIO(), io.StringIO())` — no
  subprocess, fast. Covers: usage errors, single/aggregate text, failure exit codes,
  worst-code accumulation, `--fail`, JSON schema, JSON ordering, JSON error grouping.

## Makefile changes

- `py-test`: added `PYTHONPATH=$(PY_DIR)` so tests can `import latprobe`; removed
  `|| true` now that real tests exist.