pnf

A pure-Python decoder for Shearwater's Petrel Native Format (PNF), the on-device log format used by Shearwater's Petrel-family dive computers (Petrel, Perdix, Teric, Nerd, Peregrine, and successors).

Shearwater Cloud can export your dives to XML, CSV, or its own SQLite database. The XML and CSV are lossy: they drop channels the computer actually recorded, including GF99, the deco ceiling, per-sample battery voltage and percentage, and the "@+5" prediction. The SQLite export keeps the real thing: each dive is stored as a compressed native-format blob. pnf decodes those blobs.

No dependencies, standard library only, Python 3.9+.

pip install pnf

Usage

Point it at a Shearwater Cloud "Export Database" file (dive_data.db):

import pnf

for log_id, dive in pnf.decode_database("dive_data.db"):
    h = dive.header
    print(f"{h['computer_model']} #{h['computer_serial']:X} dive {h['dive_number']}: "
          f"{h['max_depth_m']} m, {len(dive.samples)} samples")
    for s in dive.samples:
        if s.gf99 is not None:
            print(f"  t={s.t_s:.0f}s depth={s.depth_m}m gf99={s.gf99} "
                  f"ceiling={s.ceiling_m}m battery={s.battery_v}V")

Or decode a single sw-pnf blob you already have:

dive = pnf.decode(blob)

What you get

decode() returns a Dive with:

• header: a dict with the dive number and start time, and the dive configuration: gases, gradient factors, deco model, water density, atmospheric pressure, sample interval, computer serial, model, and firmware, the GNSS entry and exit fix (on computers that log it), and verbatim copies of the raw config records. start_time is seconds since 1970 on the local wall clock with no time zone, so datetime.fromtimestamp(t, timezone.utc) gives the local time the computer showed.
• samples: a list of Sample, one per log interval (typically every 10 s), each with depth, temperature, TTS, next-stop depth and time, NDL, average and per-cell ppO2, CNS, GF99, deco ceiling, battery voltage and percentage, gas fractions, setpoint, wireless-air-integration tank pressures, SAC, the decoded status/mode bits (including CCR solenoid-fired count), and the "@+5" prediction. Channels the computer did not record are None. Depths are in metres even on an imperial-units log.
• gases: the ten programmed gases (O2/He, open- vs closed-circuit, enabled).
• events: info-event records.
• unknown_records: every record type this decoder does not interpret, kept verbatim as (type_key, seq, raw_bytes). Shearwater's own desktop parser also leaves several record types (tissue snapshots, validity maps, surface ppO2, button pushes, DiveCAN events) undecoded; pnf preserves them rather than discarding them, so a future decoder can revisit them.

Under the DCIEM deco model, sample bytes 24-25 encode a safe-ascent depth instead of a ceiling / GF99 pair; pnf decodes them to safe_ascent_depth_m and leaves ceiling_m / gf99 as None. See SPEC.md for details.

The format

SPEC.md documents the container, the record types, and every byte of the dive-sample record, with the provenance of each field.

Provenance and scope

This is independent reverse engineering for interoperability: reading dive logs you own, in Python, without going through Shearwater Cloud. The documented sample fields follow the layout in libdivecomputer's shearwater_predator_parser.c. The remaining fields were identified by observation across real dives and cross-checked against Shearwater Cloud's own outputs (its XML export and its per-dive EndGF99 summary), then reconciled for naming and edge cases such as the DCIEM branch.

pnf is not affiliated with or endorsed by Shearwater Research, contains no Shearwater code, and does not redistribute any Shearwater software. Format facts (byte offsets and field meanings) are not code. If you want a maintained, officially-supported path, Shearwater offers a developer program.

Validated against Perdix 2 and Petrel 3 logs. The sample record layout is stable across native-format versions, but other models may populate fields this decoder has not seen. If a dive decodes wrong, open an issue with the blob; that is exactly the data that improves the decoder.

License

MIT. See LICENSE.