"""Figures for one run session, one flat folder per session (figures/<subject>/<session>/):

    overview.png          every source on one time axis -- start here
    data-<source>.png     the published data as recorded, one figure per source: imu (a mid-run window of
                          every sensor, accel + gyro), phone (GPS speed, pace vs distance, elevation, steps),
                          h10 (Polar ECG + accel), fit3 (Samsung watch HR / cadence / speed)
    derived-<view>.png    anything computed from the data: spectrum (foot-accel PSD vs Nyquist) today; the
                          running metrics (cadence, symmetry, ...) join once validated

Uniform units: acceleration in g, angular velocity in deg/s.

    from lvl_running import plot_session
    plot_session("data/S01/R01/level", "figures/S01/R01")
"""
from __future__ import annotations

import glob
import os

import numpy as np
import pandas as pd
from scipy.signal import welch

from .fit3 import load_fit3
from .session import load_session

G = 9.80665
DPS = 180.0 / np.pi
ACC_RAIL_G = 16.0                 # +/-16 g accelerometer range
# house palette (matches lvl_toolkit.graphing): axes x/y/z = red/green/blue; bilateral feet blue/green
AXIS_COLORS = ("#d62728", "#2ca02c", "#1f77b4")            # x = red, y = green, z = blue
FOOT_COLORS = {"left_foot": "#1f77b4", "right_foot": "#228B22"}
DPI = 110                          # fixed so a re-render of unchanged data is byte-stable in intent


def _plt():
    try:
        import matplotlib; matplotlib.use("Agg"); import matplotlib.pyplot as plt
        return plt
    except ImportError:
        return None


def _save(fig, plt, out_dir: str, name: str) -> str:
    os.makedirs(out_dir, exist_ok=True)
    p = os.path.join(out_dir, name)
    fig.savefig(p, dpi=DPI)
    plt.close(fig)
    return p


def _mid_window(ts: np.ndarray, window_s: float, frac: float = 0.4) -> slice:
    dur = (ts[-1] - ts[0]) / 1e6
    if dur <= window_s:
        return slice(0, len(ts))
    i0 = int(np.searchsorted(ts, ts[0] + int(dur * frac * 1e6)))
    fs = 1e6 / np.median(np.diff(ts))
    return slice(i0, i0 + int(window_s * fs))


def plot_imu(run, out_dir: str, window_s: float = 6.0) -> str | None:
    sensors = sorted(run.imus.values(), key=lambda s: s.placement)
    plt = _plt()
    if not sensors or plt is None:
        return None
    n = len(sensors)
    fig, ax = plt.subplots(n, 2, figsize=(12, 2.3 * n), sharex="col", squeeze=False)
    for r, s in enumerate(sensors):
        ts = s.df["time"].to_numpy(np.int64); w = _mid_window(ts, window_s)
        t = (ts[w] - ts[w][0]) / 1e6
        acc = s.df[["accel_x", "accel_y", "accel_z"]].to_numpy() / G           # g
        gyr = s.df[["gyro_x", "gyro_y", "gyro_z"]].to_numpy() * DPS            # deg/s
        peak_g = float(np.linalg.norm(acc, axis=1).max()); peak_dps = float(np.abs(gyr).max())
        for c, (d, lab, u, peak) in enumerate([(acc[w], "accel", "g", peak_g),
                                               (gyr[w], "gyro", "deg/s", peak_dps)]):
            a = ax[r][c]
            for k, axis in enumerate("xyz"):
                a.plot(t, d[:, k], lw=0.7, color=AXIS_COLORS[k], label=axis)
            if c == 0:   # impact highlight: accel samples reaching the +/-16 g rail (under-resolved impacts)
                mag_w = np.linalg.norm(acc[w], axis=1)
                hit = mag_w >= 0.95 * ACC_RAIL_G
                if hit.any():
                    a.plot(t[hit], mag_w[hit], "k.", ms=4)
                    a.axhline(ACC_RAIL_G, color="k", ls=":", lw=0.7)
                    a.axhline(-ACC_RAIL_G, color="k", ls=":", lw=0.7)
            a.set_ylabel(f"{s.placement}\n{lab} ({u})", fontsize=8); a.grid(alpha=.3)
            a.text(0.99, 0.03, f"max |{lab}| = {peak:.0f} {u}" if u == "deg/s" else f"max |{lab}| = {peak:.1f} {u}",
                   transform=a.transAxes, ha="right", va="bottom", fontsize=7, color="0.3")
            if r == 0:
                a.legend(ncol=3, fontsize=7, loc="upper right")
                a.set_title({0: "Accelerometer (g)", 1: "Gyroscope (deg/s)"}[c])
    # same y-limits on both feet per channel so left vs right is directly comparable
    foot_rows = [r for r, s in enumerate(sensors) if s.placement in ("left_foot", "right_foot")]
    for c in (0, 1):
        lims = [ax[r][c].get_ylim() for r in foot_rows]
        if len(lims) > 1:
            lo = min(l[0] for l in lims); hi = max(l[1] for l in lims)
            for r in foot_rows:
                ax[r][c].set_ylim(lo, hi)
    ax[-1][0].set_xlabel("time (s)"); ax[-1][1].set_xlabel("time (s)")
    fig.suptitle(f"Raw IMU -- {window_s:.0f}s mid-run window -- black = accel at the +/-16 g rail "
                 f"(under-resolved impacts)\n{run.name}", y=0.998, fontsize=10)
    fig.tight_layout()
    return _save(fig, plt, out_dir, "data-imu.png")


def plot_spectrum(run, out_dir: str) -> str | None:
    """Foot-accel power spectrum vs Nyquist -- shows where the 100 Hz rate is too low (aliasing)."""
    feet = [f for f in ("left_foot", "right_foot") if f in run.imus]
    plt = _plt()
    if not feet or plt is None:
        return None
    fig, ax = plt.subplots(figsize=(12, 4))
    nyq = None
    for foot in feet:
        s = run.imus[foot].df
        ts = s["time"].to_numpy(np.int64)
        fs = 1e6 / np.median(np.diff(ts)); nyq = fs / 2
        mag = np.linalg.norm(s[["accel_x", "accel_y", "accel_z"]].to_numpy() / G, axis=1)
        f, pxx = welch(mag - mag.mean(), fs=fs, nperseg=min(2048, len(mag)))
        ax.semilogy(f, pxx, lw=1.0, color=FOOT_COLORS.get(foot, "0.5"), label=foot.replace("_", " "))
    if nyq:
        ax.axvline(nyq, color="r", ls="--", lw=1.0, label=f"Nyquist ({nyq:.0f} Hz)")
    ax.set_xlabel("frequency (Hz)"); ax.set_ylabel("foot accel PSD (g^2/Hz)")
    ax.set_title(f"Foot-accel spectrum -- most energy < 30 Hz (100 Hz OK for timing); the tail still at "
                 f"the {nyq:.0f} Hz Nyquist = impact transient under-sampled\n{run.name}", fontsize=9)
    ax.grid(alpha=.3, which="both"); ax.legend()
    fig.tight_layout()
    return _save(fig, plt, out_dir, "derived-spectrum.png")


def plot_phone(run_dir: str, out_dir: str) -> str | None:
    """Phone streams: GPS-derived speed (coordinate-free pace_profile), barometric relative altitude and
    cumulative steps on one session-relative time axis, plus pace vs distance from the same profile."""
    pp, baro, stp = (os.path.join(run_dir, f) for f in ("pace_profile.csv", "barometer.csv", "steps.csv"))
    plt = _plt()
    if not any(os.path.exists(f) for f in (pp, baro, stp)) or plt is None:
        return None
    starts, dfs = [], {}
    if os.path.exists(pp):
        dfs["pp"] = pd.read_csv(pp); starts.append(int(dfs["pp"]["time_us"].iloc[0]))
    if os.path.exists(baro):
        dfs["baro"] = pd.read_csv(baro); starts.append(int(dfs["baro"]["timestamp"].iloc[0]))
    if os.path.exists(stp):
        dfs["stp"] = pd.read_csv(stp); starts.append(int(dfs["stp"]["timestamp"].iloc[0]))
    t0 = min(starts)
    fig = plt.figure(figsize=(12, 8))
    gs = fig.add_gridspec(4, 1)
    ax = [fig.add_subplot(gs[0])]
    ax += [fig.add_subplot(gs[i], sharex=ax[0]) for i in (1, 2)]     # three panels vs time ...
    ax.append(fig.add_subplot(gs[3]))                                 # ... and pace vs distance
    if "pp" in dfs:
        d = dfs["pp"]; t = (d["time_us"].to_numpy(np.int64) - t0) / 1e6 / 60.0
        ax[0].plot(t, d["speed_mps"], lw=0.8, color="#1f77b4")
        if "baro_alt_rel_m" in d:
            ax[1].plot(t, d["baro_alt_rel_m"], lw=0.8, color="#8c564b", label="baro (pace_profile)")
        if "gps_alt_rel_m" in d:
            ax[1].plot(t, d["gps_alt_rel_m"], lw=0.6, color="#bbbbbb", label="GPS")
    if "baro" in dfs and "pp" not in dfs:
        d = dfs["baro"]; t = (d["timestamp"].to_numpy(np.int64) - t0) / 1e6 / 60.0
        ax[1].plot(t, d["altitude_m"] - d["altitude_m"].iloc[0], lw=0.8, color="#8c564b", label="baro")
    ax[0].set_ylabel("speed (m/s)"); ax[0].set_title("Phone GPS -- ground speed (coordinate-free pace_profile)"); ax[0].grid(alpha=.3)
    ax[1].set_ylabel("rel. altitude (m)"); ax[1].set_title("Barometer -- relative altitude"); ax[1].grid(alpha=.3)
    if ax[1].get_legend_handles_labels()[0]:
        ax[1].legend(fontsize=7)
    if "stp" in dfs:
        d = dfs["stp"]; t = (d["timestamp"].to_numpy(np.int64) - t0) / 1e6 / 60.0
        ax[2].plot(t, d["step_count"], lw=0.9, color="#2ca02c")
    ax[2].set_ylabel("cumulative steps"); ax[2].set_title("Phone step counter"); ax[2].set_xlabel("time (min)"); ax[2].grid(alpha=.3)
    if "pp" in dfs and len(dfs["pp"]) > 1:                       # the same profile against distance
        prof = dfs["pp"]
        d = prof["cum_km"].to_numpy() - float(prof["cum_km"].iloc[0])
        ax[3].plot(d, prof["pace_min_km"].to_numpy(), lw=0.5, color="0.75", label="pace per GPS fix")
        ax[3].plot(d, prof["pace_min_km"].rolling(15, center=True, min_periods=1).median().to_numpy(),
                   lw=1.5, color="#377eb8", label="rolling median")
        ax[3].set_ylim(3, 12); ax[3].invert_yaxis(); ax[3].legend(fontsize=8, loc="lower right")
    ax[3].set_ylabel("pace (min/km)"); ax[3].set_title("Phone GPS -- pace vs distance"); ax[3].set_xlabel("distance (km)"); ax[3].grid(alpha=.3)
    d = os.path.normpath(run_dir)        # same label as Session.name: R01/level
    fig.suptitle(f"Phone streams -- {os.path.basename(os.path.dirname(d))}/{os.path.basename(d)}", y=0.999, fontsize=10)
    fig.tight_layout()
    return _save(fig, plt, out_dir, "data-phone.png")


def plot_h10(run_dir: str, out_dir: str, ecg_s: float = 4.0, acc_s: float = 6.0) -> str | None:
    """Polar H10 chest strap: a mid-run window of the ECG waveform and the strap accelerometer."""
    ecgf = glob.glob(os.path.join(run_dir, "polar_*_ecg.csv"))
    accf = glob.glob(os.path.join(run_dir, "polar_*_acc.csv"))
    plt = _plt()
    if (not ecgf and not accf) or plt is None:
        return None

    def mid(rx, span_s):
        t0 = rx[0] + (rx[-1] - rx[0]) * 0.4
        m = (rx >= t0) & (rx <= t0 + span_s * 1e6)
        return m, (rx[m] - (rx[m][0] if m.any() else rx[0])) / 1e6

    fig, ax = plt.subplots(2, 1, figsize=(12, 5))
    if ecgf:
        e = pd.read_csv(ecgf[0]); rx = e["RECEIVED_TIMESTAMP"].to_numpy(float)
        m, t = mid(rx, ecg_s)
        ax[0].plot(t, e["ECG"].to_numpy()[m], lw=0.6, color="#c00")
    ax[0].set_ylabel("ECG (uV)"); ax[0].set_title("Polar H10 -- ECG"); ax[0].grid(alpha=.3)
    if accf:
        a = pd.read_csv(accf[0]); rx = a["RECEIVED_TIMESTAMP"].to_numpy(float)
        m, t = mid(rx, acc_s)
        cols = [c for c in ("ACCEL_X", "ACCEL_Y", "ACCEL_Z") if c in a]
        colors = {"x": AXIS_COLORS[0], "y": AXIS_COLORS[1], "z": AXIS_COLORS[2]}
        for c in cols:
            ax[1].plot(t, a[c].to_numpy()[m] / 1000.0, lw=0.6, color=colors.get(c[-1].lower()), label=c[-1].lower())
        if cols:
            ax[1].legend(fontsize=7)
    ax[1].set_ylabel("ACC (g)"); ax[1].set_xlabel("time (s)")
    ax[1].set_title("Polar H10 -- accelerometer (g)"); ax[1].grid(alpha=.3)
    fig.tight_layout()
    return _save(fig, plt, out_dir, "data-h10.png")


def plot_fit3(run_dir: str, out_dir: str) -> str | None:
    """Samsung Fit3 watch (non-LEVEL reference): heart rate, cadence and speed over the workout."""
    f = load_fit3(os.path.dirname(os.path.normpath(run_dir)))      # fit3/ is at the day level
    plt = _plt()
    if f is None or f.live.empty or plt is None:
        return None
    live = f.live
    t = (live["time"].to_numpy(float) - float(live["time"].iloc[0])) / 60e6    # us -> min
    col = lambda c: live[c].to_numpy(float) if c in live else np.full(len(live), np.nan)
    fig, ax = plt.subplots(3, 1, figsize=(12, 6), sharex=True)
    ax[0].plot(t, col("heart_rate"), lw=0.9, color="#c00"); ax[0].set_ylabel("HR (bpm)")
    ax[0].set_title("Fit3 -- heart rate"); ax[0].grid(alpha=.3)
    ax[1].plot(t, col("cadence"), lw=0.9, color="#ff7f0e"); ax[1].set_ylabel("cadence (spm)")
    ax[1].set_title("Fit3 -- cadence"); ax[1].grid(alpha=.3)
    ax[2].plot(t, col("speed"), lw=0.9, color="#1f77b4"); ax[2].set_ylabel("speed (m/s)")
    ax[2].set_title("Fit3 -- speed"); ax[2].set_xlabel("time (min)"); ax[2].grid(alpha=.3)
    fig.suptitle("Non-LEVEL reference -- Samsung Galaxy Fit3 watch", y=0.999, fontsize=10)
    fig.tight_layout()
    return _save(fig, plt, out_dir, "data-fit3.png")


PLACEMENT_COLORS = {**FOOT_COLORS, "back": "#9467bd", "left_wrist": "#ff7f0e", "right_wrist": "#d62728",
                    "wrist": "#ff7f0e", "arm": "#8c564b"}


def _binned_std(t_us: np.ndarray, x: np.ndarray, t0: int, bin_s: float) -> tuple[np.ndarray, np.ndarray]:
    """Per-bin standard deviation of x (NaN where a bin has no samples -> dropouts show as gaps)."""
    i = ((t_us - t0) // int(bin_s * 1e6)).astype(np.int64)
    keep = i >= 0
    i, x = i[keep], x[keep]
    n = np.bincount(i).astype(float)
    s1, s2 = np.bincount(i, x), np.bincount(i, x * x)
    with np.errstate(invalid="ignore", divide="ignore"):
        sd = np.sqrt(np.maximum(s2 / n - (s1 / n) ** 2, 0))
    sd[n == 0] = np.nan
    return np.arange(len(n)) * bin_s / 60.0, sd


def plot_overview(run_dir: str, out_dir: str) -> str | None:
    """The one-look summary of a run: every source on one time axis -- LEVEL sensor movement, heart rate
    (chest strap + watch), speed (phone GPS + watch), cadence (phone steps + watch), relative elevation --
    titled with the setup. Dropouts show as gaps."""
    from .catalog import session_row
    plt = _plt()
    run = load_session(run_dir)
    if plt is None or not run.imus:
        return None
    root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.normpath(run_dir))))
    rr = sr = session_row(run_dir, root)
    t0 = min(int(s.df["time"].iloc[0]) for s in run.imus.values())
    t_end = max(int(s.df["time"].iloc[-1]) for s in run.imus.values())
    tmin = lambda us: (np.asarray(us, dtype=float) - t0) / 60e6
    read = lambda name: pd.read_csv(p) if (p := next(iter(sorted(glob.glob(os.path.join(run_dir, name)))), None)) else None
    fit3 = load_fit3(os.path.dirname(os.path.normpath(run_dir)))
    live = fit3.live if fit3 is not None and not fit3.live.empty else None
    pace, steps, hr = read("pace_profile.csv"), read("steps.csv"), read("polar_*_hr.csv")

    fig, ax = plt.subplots(5, 1, figsize=(12, 11), sharex=True)
    none = lambda a: a.text(0.5, 0.5, "not recorded", transform=a.transAxes, ha="center", va="center", color="0.5")

    # 1 movement: 1 s variability of |accel| per sensor
    for pl, s in sorted(run.imus.items()):
        mag = np.linalg.norm(s.df[["accel_x", "accel_y", "accel_z"]].to_numpy() / G, axis=1)
        x, sd = _binned_std(s.df["time"].to_numpy(np.int64), mag, t0, 1.0)
        ax[0].plot(x, sd, lw=0.7, color=PLACEMENT_COLORS.get(pl, "0.4"), label=pl.replace("_", " "))
    ax[0].set_ylabel("movement\n(1 s SD of |accel|, g)"); ax[0].legend(ncol=len(run.imus), fontsize=8, loc="upper right")
    ax[0].set_title("LEVEL sensors -- movement intensity (gaps = no data)")

    # 2 heart rate
    if hr is not None and len(hr):
        h = hr[hr["HEART_RATE"] > 0]
        ax[1].plot(tmin(h["RECEIVED_TIMESTAMP"]), h["HEART_RATE"], lw=0.9, color="#c00", label="Polar H10 (chest)")
    if live is not None and "heart_rate" in live:
        ax[1].plot(tmin(live["time"]), live["heart_rate"], lw=1.2, ls="--", color="#ff7f0e", label="Fit3 (wrist)")
    ax[1].set_ylabel("heart rate\n(bpm)"); ax[1].set_title("Heart rate")
    ax[1].legend(fontsize=8, loc="lower right") if ax[1].lines else none(ax[1])

    # 3 speed
    if pace is not None and "speed_mps" in pace:
        ax[2].plot(tmin(pace["time_us"]), pace["speed_mps"].rolling(15, center=True, min_periods=1).median(),
                   lw=0.9, color="#1f77b4", label="phone GPS")
    if live is not None and "speed" in live:
        ax[2].plot(tmin(live["time"]), live["speed"], lw=1.2, ls="--", color="#ff7f0e", label="Fit3")
    ax[2].set_ylabel("speed\n(m/s)"); ax[2].set_title("Speed")
    ax[2].legend(fontsize=8, loc="lower right") if ax[2].lines else none(ax[2])

    # 4 cadence: phone step events per 30 s (x2 -> steps/min) + watch
    if steps is not None and len(steps):
        ts = steps["timestamp"].to_numpy(np.int64)
        i = (ts - t0) // int(30e6); i = i[i >= 0]
        if len(i):
            n = np.bincount(i).astype(float); n[n == 0] = np.nan
            ax[3].step(np.arange(len(n)) * 0.5, n * 2, where="post", lw=0.9, color="#2ca02c", label="phone step counter")
    if live is not None and "cadence" in live:
        ax[3].plot(tmin(live["time"]), live["cadence"], lw=1.2, ls="--", color="#ff7f0e", label="Fit3")
    ax[3].set_ylabel("cadence\n(steps/min)"); ax[3].set_title("Cadence")
    ax[3].legend(fontsize=8, loc="lower right") if ax[3].lines else none(ax[3])

    # 5 relative elevation
    if pace is not None and "baro_alt_rel_m" in pace:
        ax[4].plot(tmin(pace["time_us"]), pace["baro_alt_rel_m"], lw=0.9, color="#8c564b", label="phone barometer")
    if pace is not None and "gps_alt_rel_m" in pace:
        ax[4].plot(tmin(pace["time_us"]), pace["gps_alt_rel_m"], lw=0.6, color="#bbbbbb", label="phone GPS")
    ax[4].set_ylabel("relative\nelevation (m)"); ax[4].set_title("Elevation (relative)")
    ax[4].legend(fontsize=8, loc="lower right") if ax[4].lines else none(ax[4])

    for a in ax:
        a.grid(alpha=.3)
    ax[-1].set_xlabel("time since the first LEVEL sample (min)")
    # whole session, not just the LEVEL span: sensors that stop early show as lines ending early
    ax[-1].set_xlim(-0.5, max((t_end - t0) / 60e6, rr.get("duration_min") or 0) + 0.5)
    fmt = lambda v, f: "?" if v is None or (isinstance(v, float) and np.isnan(v)) else f.format(v)
    refs = ", ".join(k for k, ok in (("Polar H10", sr.get("polar_h10")), ("Fit3 watch", sr.get("fit3_watch"))) if ok)
    where = " ".join(v for v in (sr.get("setting"), sr.get("surface")) if v) or "?"   # no event / place: figures stay anonymous
    fig.suptitle(f"{run.name}  --  {where}, {fmt(rr.get('duration_min'), '{:.0f}')} min, "
                 f"{fmt(rr.get('distance_km'), '{:.2f}')} km\n"
                 f"{sr.get('sensor_count')} LEVEL sensors ({sr.get('placements', '').replace(';', ', ')}) via "
                 f"{sr.get('transport') or '?'} at {sr.get('sampling_rate_hz') or '?'} Hz  |  references: {refs or 'none'}",
                 fontsize=11)
    fig.tight_layout()
    return _save(fig, plt, out_dir, "overview.png")


def plot_session(run_dir: str, out_dir: str) -> dict[str, str | None]:
    """Every figure for one run session into out_dir (flat: overview.png, data-*.png, derived-*.png).
    Returns {figure: path or None}."""
    run = load_session(run_dir)
    return {"overview": plot_overview(run_dir, out_dir),
            "data-imu": plot_imu(run, out_dir), "data-phone": plot_phone(run_dir, out_dir),
            "data-h10": plot_h10(run_dir, out_dir), "data-fit3": plot_fit3(run_dir, out_dir),
            "derived-spectrum": plot_spectrum(run, out_dir)}