app.py

#!/usr/bin/env python3
"""Streamlit RTSS diff viewer app."""

from __future__ import annotations

import sys
import tempfile
from math import sqrt
from pathlib import Path
from typing import Any

import plotly.graph_objects as go
import streamlit as st

ROOT = Path(__file__).resolve().parent
SRC_DIR = ROOT / "src"
if str(SRC_DIR) not in sys.path:
    sys.path.insert(0, str(SRC_DIR))

from rtssdiffviewer.dcm_to_json import dcm_to_json  # noqa: E402
from rtssdiffviewer.diff_core import (  # noqa: E402
    COMPONENT_KEYS,
    DEFAULT_VOLATILE_TAG_PREFIXES,
    contour_diff_text,
    get_contour_slices_structured,
    normalize_value,
    pretty_json_text,
    select_component,
    unified_diff_text,
)

try:
    from st_diff_viewer import diff_viewer
except Exception:
    diff_viewer = None


def dcm_bytes_to_json_dict(payload: bytes) -> dict[str, Any]:
    with tempfile.NamedTemporaryFile(suffix=".dcm", delete=False) as tmp:
        tmp.write(payload)
        tmp_path = Path(tmp.name)
    try:
        return dcm_to_json(tmp_path)
    finally:
        tmp_path.unlink(missing_ok=True)


def ensure_state() -> None:
    st.session_state.setdefault("left_name", None)
    st.session_state.setdefault("right_name", None)
    st.session_state.setdefault("left_json_raw", None)
    st.session_state.setdefault("right_json_raw", None)
    st.session_state.setdefault("batch_variants", {})
    st.session_state.setdefault("pair_step", "Upload Pair")
    st.session_state.setdefault("pair_diff_requested", False)
    st.session_state.setdefault("pair_diff_sig", None)
    st.session_state.setdefault("pair_focus_slice_z", None)
    st.session_state.setdefault("axial_target_slice_z", None)
    st.session_state.setdefault("contour_detail_target_slice_z", None)


def _point_distance(a: tuple[float, float, float], b: tuple[float, float, float]) -> float:
    return sqrt((a[0] - b[0]) ** 2 + (a[1] - b[1]) ** 2 + (a[2] - b[2]) ** 2)


def _greedy_point_matches(
    left_points: list[tuple[float, float, float]],
    right_points: list[tuple[float, float, float]],
) -> list[tuple[int, int, float]]:
    if not left_points or not right_points:
        return []

    candidates: list[tuple[float, int, int]] = []
    for li, lp in enumerate(left_points):
        for ri, rp in enumerate(right_points):
            candidates.append((_point_distance(lp, rp), li, ri))
    candidates.sort(key=lambda x: x[0])

    used_left: set[int] = set()
    used_right: set[int] = set()
    matches: list[tuple[int, int, float]] = []
    for dist, li, ri in candidates:
        if li in used_left or ri in used_right:
            continue
        used_left.add(li)
        used_right.add(ri)
        matches.append((li, ri, dist))
    return matches


def _slice_point_counts(
    left_rois: dict[str, list[tuple[float, float, float]]],
    right_rois: dict[str, list[tuple[float, float, float]]],
) -> tuple[int, int, list[tuple[float, float, float]], list[tuple[float, float, float]]]:
    left_points = [p for pts in left_rois.values() for p in pts]
    right_points = [p for pts in right_rois.values() for p in pts]
    return len(left_points), len(right_points), left_points, right_points


def _slice_match_metrics(
    left_rois: dict[str, list[tuple[float, float, float]]],
    right_rois: dict[str, list[tuple[float, float, float]]],
    tolerance_mm: float,
) -> dict[str, Any]:
    left_count, right_count, left_points, right_points = _slice_point_counts(left_rois, right_rois)
    matches = _greedy_point_matches(left_points, right_points)
    matched_in_tol = sum(1 for _, _, dist in matches if dist <= tolerance_mm)
    dice = (2.0 * matched_in_tol) / (left_count + right_count) if (left_count + right_count) > 0 else 1.0
    mismatch_count = left_count + right_count - (2 * matched_in_tol)
    count_delta = abs(left_count - right_count)

    left_roi_names = set(left_rois.keys())
    right_roi_names = set(right_rois.keys())
    all_roi_names = sorted(left_roi_names | right_roi_names)
    identical_slice = left_roi_names == right_roi_names and all(
        sorted(left_rois.get(roi, [])) == sorted(right_rois.get(roi, []))
        for roi in all_roi_names
    )

    return {
        "left_count": left_count,
        "right_count": right_count,
        "dice": dice,
        "mismatch_count": mismatch_count,
        "count_delta": count_delta,
        "left_roi_names": left_roi_names,
        "right_roi_names": right_roi_names,
        "all_roi_names": all_roi_names,
        "identical_slice": identical_slice,
    }


def _safe_key_fragment(value: str) -> str:
    out = []
    for ch in value:
        out.append(ch if ch.isalnum() else "_")
    return "".join(out)


def _nearest_slice_value(slices: list[float], target: float | None) -> float | None:
    if not slices:
        return None
    if target is None:
        return slices[0]
    return min(slices, key=lambda z: (abs(z - target), z))


def _step_slice_value(slices: list[float], current: float, step: int) -> float:
    if not slices:
        return current
    try:
        idx = slices.index(current)
    except ValueError:
        idx = 0
    next_idx = max(0, min(len(slices) - 1, idx + step))
    return slices[next_idx]


def _format_slice_rois_text(rois: dict[str, list[tuple[float, float, float]]], precision: int) -> str:
    if not rois:
        return "(no contours on this slice)"

    lines: list[str] = []
    for roi_name in sorted(rois.keys()):
        points = rois[roi_name]
        lines.append(f"{roi_name}: {len(points)} points")
        for i, (x, y, z) in enumerate(points, start=1):
            lines.append(f"  {i:03d}: ({x:.{precision}f}, {y:.{precision}f}, {z:.{precision}f})")
        lines.append("")
    return "\n".join(lines).strip()


def _extract_ordered_contours_by_slice(
    rtss_json: dict[str, Any],
    precision: int = 4,
) -> dict[float, list[dict[str, Any]]]:
    """Extract ordered contour polylines grouped by axial slice (z)."""
    slices: dict[float, list[dict[str, Any]]] = {}
    roi_contour_seq = rtss_json.get("(3006,0039) ROIContourSequence", [])
    if not isinstance(roi_contour_seq, list):
        return slices

    for roi_idx, roi_item in enumerate(roi_contour_seq):
        if not isinstance(roi_item, dict):
            continue

        roi_number = roi_item.get("(3006,0084) ReferencedROINumber")
        if roi_number is None:
            roi_number = roi_idx
        roi_name = f"ROI {roi_number}"

        contour_seq = roi_item.get("(3006,0040) ContourSequence", [])
        if not isinstance(contour_seq, list):
            continue

        for contour_idx, contour_item in enumerate(contour_seq, start=1):
            if not isinstance(contour_item, dict):
                continue

            points = _extract_xyz_points(contour_item.get("(3006,0050) ContourData", []))
            if not points:
                continue

            z_mean = sum(p[2] for p in points) / len(points)
            z_key = round(z_mean, precision)
            contour_number = contour_item.get("(3006,0048) ContourNumber", contour_idx)

            slices.setdefault(z_key, []).append(
                {
                    "roi_name": roi_name,
                    "contour_label": f"{roi_name} | Contour {contour_number}",
                    "points": points,
                }
            )

    return slices


def _add_direction_annotation(
    fig: go.Figure,
    points: list[tuple[float, float, float]],
    color: str,
    text: str,
) -> None:
    if len(points) < 2:
        return
    x0, y0, _ = points[0]
    x1, y1, _ = points[1]
    fig.add_annotation(
        x=x1,
        y=y1,
        ax=x0,
        ay=y0,
        xref="x",
        yref="y",
        axref="x",
        ayref="y",
        showarrow=True,
        arrowhead=3,
        arrowsize=1,
        arrowwidth=1.6,
        arrowcolor=color,
        text=text,
        font={"size": 11, "color": color},
        align="left",
        xanchor="left",
    )


def _add_contour_traces_2d(
    fig: go.Figure,
    contours: list[dict[str, Any]],
    side_name: str,
    line_color: str,
    line_dash: str,
) -> None:
    for idx, contour in enumerate(contours, start=1):
        points = contour["points"]
        if not points:
            continue

        xs = [p[0] for p in points]
        ys = [p[1] for p in points]
        point_ids = list(range(1, len(points) + 1))
        contour_name = contour["contour_label"]

        fig.add_trace(
            go.Scatter(
                x=xs,
                y=ys,
                mode="lines+markers",
                line={"width": 2, "color": line_color, "dash": line_dash},
                marker={"size": 5, "color": line_color},
                name=f"{side_name}: {contour_name}",
                legendgroup=f"{side_name}_{idx}",
                hovertemplate=(
                    f"{side_name}<br>{contour_name}<br>Point %{{customdata}}"
                    "<br>x=%{x:.3f}<br>y=%{y:.3f}<extra></extra>"
                ),
                customdata=point_ids,
            )
        )

        first = points[0]
        last = points[-1]
        fig.add_trace(
            go.Scatter(
                x=[first[0]],
                y=[first[1]],
                mode="markers+text",
                marker={"size": 11, "color": line_color, "symbol": "star"},
                text=["Start"],
                textposition="top center",
                name=f"{side_name} start",
                legendgroup=f"{side_name}_{idx}",
                showlegend=False,
                hovertemplate=(
                    f"{side_name}<br>{contour_name}<br>Start point (index 1)"
                    "<br>x=%{x:.3f}<br>y=%{y:.3f}<extra></extra>"
                ),
            )
        )
        fig.add_trace(
            go.Scatter(
                x=[last[0]],
                y=[last[1]],
                mode="markers+text",
                marker={"size": 10, "color": line_color, "symbol": "x"},
                text=[f"End ({len(points)})"],
                textposition="bottom center",
                name=f"{side_name} end",
                legendgroup=f"{side_name}_{idx}",
                showlegend=False,
                hovertemplate=(
                    f"{side_name}<br>{contour_name}<br>End point (index {len(points)})"
                    "<br>x=%{x:.3f}<br>y=%{y:.3f}<extra></extra>"
                ),
            )
        )

        _add_direction_annotation(fig, points, line_color, f"{side_name} direction")


def render_axial_contour_view(
    *,
    left_name: str,
    right_name: str,
    left_raw: dict[str, Any],
    right_raw: dict[str, Any],
    precision: int = 4,
) -> None:
    left_by_slice = _extract_ordered_contours_by_slice(left_raw, precision=precision)
    right_by_slice = _extract_ordered_contours_by_slice(right_raw, precision=precision)
    all_slices = sorted(set(left_by_slice.keys()) | set(right_by_slice.keys()))

    if not all_slices:
        st.warning("No contour data found in either RTSS file.")
        return

    summary_left = sum(len(v) for v in left_by_slice.values())
    summary_right = sum(len(v) for v in right_by_slice.values())
    only_left = [z for z in all_slices if z in left_by_slice and z not in right_by_slice]
    only_right = [z for z in all_slices if z in right_by_slice and z not in left_by_slice]

    m1, m2, m3, m4 = st.columns(4)
    with m1:
        st.metric(f"{left_name} slices", f"{len(left_by_slice):,}")
    with m2:
        st.metric(f"{right_name} slices", f"{len(right_by_slice):,}")
    with m3:
        st.metric(f"{left_name} contours", f"{summary_left:,}")
    with m4:
        st.metric(f"{right_name} contours", f"{summary_right:,}")

    if only_left or only_right:
        notes: list[str] = []
        if only_left:
            notes.append(f"Slices only in {left_name}: {len(only_left)}")
        if only_right:
            notes.append(f"Slices only in {right_name}: {len(only_right)}")
        st.warning(" | ".join(notes))
    else:
        st.success("Both files have contours on the same set of axial slice positions.")

    target_z = st.session_state.axial_target_slice_z
    if target_z is not None:
        aligned_target = _nearest_slice_value(all_slices, target_z)
        if aligned_target is not None:
            st.session_state.axial_slice_select = aligned_target
        st.session_state.axial_target_slice_z = None
    elif st.session_state.get("axial_slice_select") not in all_slices:
        st.session_state.axial_slice_select = all_slices[0]

    nav_col_1, nav_col_2, nav_col_3 = st.columns(3)
    with nav_col_1:
        if st.button("Previous Slice", key="axial_prev_slice"):
            st.session_state.axial_slice_select = _step_slice_value(
                all_slices,
                st.session_state.get("axial_slice_select", all_slices[0]),
                -1,
            )
            st.rerun()
    with nav_col_2:
        if st.button("Next Slice", key="axial_next_slice"):
            st.session_state.axial_slice_select = _step_slice_value(
                all_slices,
                st.session_state.get("axial_slice_select", all_slices[0]),
                1,
            )
            st.rerun()
    with nav_col_3:
        if st.button("Open Contour Detail View", key="axial_open_contour_detail"):
            current_z = st.session_state.get("axial_slice_select", all_slices[0])
            st.session_state.contour_detail_target_slice_z = current_z
            st.session_state.pair_focus_slice_z = current_z
            st.session_state.pair_step = "Contour Detail View"
            st.rerun()

    selected_z = st.select_slider(
        "Axial slice z (mm)",
        options=all_slices,
        format_func=lambda z: f"{z:.{precision}f}",
        value=st.session_state.get("axial_slice_select", all_slices[0]),
        key="axial_slice_select",
    )
    st.session_state.pair_focus_slice_z = selected_z

    left_contours = left_by_slice.get(selected_z, [])
    right_contours = right_by_slice.get(selected_z, [])

    left_labels = {c["contour_label"] for c in left_contours}
    right_labels = {c["contour_label"] for c in right_contours}
    only_left_contours = sorted(left_labels - right_labels)
    only_right_contours = sorted(right_labels - left_labels)

    if left_contours and not right_contours:
        st.error(f"Slice z={selected_z:.{precision}f} is present only in {left_name}.")
    elif right_contours and not left_contours:
        st.error(f"Slice z={selected_z:.{precision}f} is present only in {right_name}.")
    else:
        st.info(
            f"Slice z={selected_z:.{precision}f} has contours in both files: "
            f"{len(left_contours)} vs {len(right_contours)}"
        )

    if only_left_contours or only_right_contours:
        mismatch_notes: list[str] = []
        if only_left_contours:
            mismatch_notes.append(f"Contours only in {left_name}: {len(only_left_contours)}")
        if only_right_contours:
            mismatch_notes.append(f"Contours only in {right_name}: {len(only_right_contours)}")
        st.warning(" | ".join(mismatch_notes))

    fig = go.Figure()
    if left_contours:
        _add_contour_traces_2d(
            fig,
            contours=left_contours,
            side_name=left_name,
            line_color="#d62728",
            line_dash="solid",
        )
    if right_contours:
        _add_contour_traces_2d(
            fig,
            contours=right_contours,
            side_name=right_name,
            line_color="#2ca02c",
            line_dash="dot",
        )

    if not fig.data:
        st.info("No contours to display for this slice.")
        return

    fig.update_layout(
        title=f"Axial Contour View at z={selected_z:.{precision}f} mm",
        xaxis_title="X (mm)",
        yaxis_title="Y (mm)",
        xaxis={"fixedrange": False},
        yaxis={"scaleanchor": "x", "scaleratio": 1, "fixedrange": False},
        dragmode="zoom",
        margin={"l": 10, "r": 10, "t": 48, "b": 10},
        legend={"orientation": "h", "y": 1.02, "x": 0},
    )
    st.plotly_chart(
        fig,
        use_container_width=True,
        config={
            "scrollZoom": True,
            "displaylogo": False,
            "modeBarButtonsToAdd": ["zoom2d", "pan2d", "resetScale2d"],
        },
    )

    with st.expander("Slice contour details", expanded=False):
        left_col, right_col = st.columns(2)
        with left_col:
            st.markdown(f"**{left_name}**")
            if not left_contours:
                st.caption("No contours on this slice.")
            else:
                for c in left_contours:
                    pts = c["points"]
                    st.write(
                        f"{c['contour_label']}: {len(pts)} points | "
                        f"start=({pts[0][0]:.{precision}f}, {pts[0][1]:.{precision}f}) | "
                        f"end=({pts[-1][0]:.{precision}f}, {pts[-1][1]:.{precision}f})"
                    )


def render_contour_detail_text_view(
    *,
    left_name: str,
    right_name: str,
    left_raw: dict[str, Any],
    right_raw: dict[str, Any],
    precision: int = 4,
) -> None:
    left_contours = select_component(left_raw, "contours")
    right_contours = select_component(right_raw, "contours")
    left_slices, right_slices = get_contour_slices_structured(left_contours, right_contours, precision=precision)
    all_slices = sorted(set(left_slices.keys()) | set(right_slices.keys()))

    if not all_slices:
        st.warning("No contour data found in either file.")
        return

    target_z = st.session_state.contour_detail_target_slice_z
    if target_z is not None:
        aligned_target = _nearest_slice_value(all_slices, target_z)
        if aligned_target is not None:
            st.session_state.contour_detail_slice_select = aligned_target
        st.session_state.contour_detail_target_slice_z = None
    elif st.session_state.get("contour_detail_slice_select") not in all_slices:
        st.session_state.contour_detail_slice_select = all_slices[0]

    nav_col_1, nav_col_2, nav_col_3 = st.columns(3)
    with nav_col_1:
        if st.button("Previous Slice", key="detail_prev_slice"):
            st.session_state.contour_detail_slice_select = _step_slice_value(
                all_slices,
                st.session_state.get("contour_detail_slice_select", all_slices[0]),
                -1,
            )
            st.rerun()
    with nav_col_2:
        if st.button("Next Slice", key="detail_next_slice"):
            st.session_state.contour_detail_slice_select = _step_slice_value(
                all_slices,
                st.session_state.get("contour_detail_slice_select", all_slices[0]),
                1,
            )
            st.rerun()
    with nav_col_3:
        if st.button("Back To Visual Comparison", key="detail_back_to_visual"):
            current_z = st.session_state.get("contour_detail_slice_select", all_slices[0])
            st.session_state.axial_target_slice_z = current_z
            st.session_state.pair_focus_slice_z = current_z
            st.session_state.pair_step = "Pair Diff"
            st.session_state.pair_component = "contours"
            st.session_state.pair_diff_requested = True
            st.session_state.pair_diff_sig = "force"
            st.rerun()

    selected_z = st.select_slider(
        "Contour slice z (mm)",
        options=all_slices,
        format_func=lambda z: f"{z:.{precision}f}",
        value=st.session_state.get("contour_detail_slice_select", all_slices[0]),
        key="contour_detail_slice_select",
    )
    st.session_state.pair_focus_slice_z = selected_z

    left_rois = left_slices.get(selected_z, {})
    right_rois = right_slices.get(selected_z, {})

    st.markdown(f"### Contour Text Comparison At z={selected_z:.{precision}f}")
    col_1, col_2 = st.columns(2)
    with col_1:
        st.markdown(f"**{left_name}**")
        st.code(_format_slice_rois_text(left_rois, precision), language="text")
    with col_2:
        st.markdown(f"**{right_name}**")
        st.code(_format_slice_rois_text(right_rois, precision), language="text")

    left_text = _format_slice_rois_text(left_rois, precision)
    right_text = _format_slice_rois_text(right_rois, precision)
    diff_text = unified_diff_text(left_text, right_text, left_name, right_name)
    st.markdown("#### Slice Unified Diff")
    if diff_text.strip():
        st.code(diff_text, language="diff")
    else:
        st.success("No textual differences on this slice.")
        with right_col:
            st.markdown(f"**{right_name}**")
            if not right_contours:
                st.caption("No contours on this slice.")
            else:
                for c in right_contours:
                    pts = c["points"]
                    st.write(
                        f"{c['contour_label']}: {len(pts)} points | "
                        f"start=({pts[0][0]:.{precision}f}, {pts[0][1]:.{precision}f}) | "
                        f"end=({pts[-1][0]:.{precision}f}, {pts[-1][1]:.{precision}f})"
                    )


def render_diff_panel(
    *,
    left_name: str,
    right_name: str,
    left_raw: dict[str, Any],
    right_raw: dict[str, Any],
    component: str,
    precision: int,
    keep_volatile: bool,
    allow_rich_view: bool,
    max_rich_chars: int,
    max_rich_lines: int,
    key_prefix: str,
) -> None:
    ignore_prefixes: set[str] = set()
    if not keep_volatile:
        ignore_prefixes.update(DEFAULT_VOLATILE_TAG_PREFIXES)

    # Get selected components
    left_selected = select_component(left_raw, component)
    right_selected = select_component(right_raw, component)

    # Check if components are identical
    if left_selected == right_selected:
        st.info(f"✅ **{component.capitalize()} components are identical** — no differences to compare.")
        col1, col2 = st.columns(2)
        with col1:
            left_json_text = pretty_json_text(left_selected)
            st.download_button(
                "Download left JSON",
                data=left_json_text,
                file_name=f"{Path(left_name).stem}.{component}.json",
                mime="application/json",
                key=f"{key_prefix}_dl_left",
            )
        with col2:
            right_json_text = pretty_json_text(right_selected)
            st.download_button(
                "Download right JSON",
                data=right_json_text,
                file_name=f"{Path(right_name).stem}.{component}.json",
                mime="application/json",
                key=f"{key_prefix}_dl_right",
            )
        return

    # Special handling for contour diffing with two-panel layout
    if component == "contours":
        left_slices, right_slices = get_contour_slices_structured(left_selected, right_selected, precision=precision)

        col1, col2, col3 = st.columns(3)
        with col1:
            left_json_text = pretty_json_text(left_selected)
            st.download_button(
                "Download left JSON",
                data=left_json_text,
                file_name=f"{Path(left_name).stem}.{component}.json",
                mime="application/json",
                key=f"{key_prefix}_dl_left",
            )
        with col2:
            right_json_text = pretty_json_text(right_selected)
            st.download_button(
                "Download right JSON",
                data=right_json_text,
                file_name=f"{Path(right_name).stem}.{component}.json",
                mime="application/json",
                key=f"{key_prefix}_dl_right",
            )
        with col3:
            diff_text = contour_diff_text(left_selected, right_selected, left_name, right_name, precision=precision)
            st.download_button(
                "Download contour diff",
                data=diff_text,
                file_name=f"{Path(left_name).stem}__{Path(right_name).stem}.{component}.diff",
                mime="text/plain",
                key=f"{key_prefix}_dl_diff",
            )

        st.markdown("### Contour Diff View (by Slice Plane)")
        st.caption(
            "Contour points are compared by slice plane (z-coordinate). "
            "Left panel shows the first file, right panel shows the second file."
        )

        if not left_slices and not right_slices:
            st.warning("No contour data found in either file.")
            return

        global_tolerance = st.slider(
            "Correspondence tolerance (mm)",
            min_value=0.1,
            max_value=10.0,
            value=1.0,
            step=0.1,
            key=f"{key_prefix}_corr_tol",
        )

        filter_col_1, filter_col_2, filter_col_3, filter_col_4 = st.columns(4)
        with filter_col_1:
            hide_identical_slices = st.checkbox(
                "Hide identical slices",
                value=False,
                key=f"{key_prefix}_hide_identical_slices",
            )
        with filter_col_2:
            max_dice_filter = st.slider(
                "Max dice to include",
                min_value=0.0,
                max_value=1.0,
                value=1.0,
                step=0.01,
                key=f"{key_prefix}_max_dice",
            )
        with filter_col_3:
            min_mismatch_filter = st.number_input(
                "Min mismatch",
                min_value=0,
                value=0,
                step=1,
                key=f"{key_prefix}_min_mismatch",
            )
        with filter_col_4:
            sort_mode = st.selectbox(
                "Sort slices",
                options=[
                    "Worst dice first",
                    "Best dice first",
                    "Highest mismatch first",
                    "Slice z ascending",
                    "Slice z descending",
                ],
                index=0,
                key=f"{key_prefix}_slice_sort",
            )

        # Build slice metrics first, then filter/sort.
        all_z_coords = sorted(set(left_slices.keys()) | set(right_slices.keys()))
        slice_rows: list[dict[str, Any]] = []
        for z in all_z_coords:
            left_rois = left_slices.get(z, {})
            right_rois = right_slices.get(z, {})
            metrics = _slice_match_metrics(left_rois, right_rois, global_tolerance)
            slice_rows.append(
                {
                    "z": z,
                    "left_rois": left_rois,
                    "right_rois": right_rois,
                    "left_count": metrics["left_count"],
                    "right_count": metrics["right_count"],
                    "dice": metrics["dice"],
                    "mismatch_count": metrics["mismatch_count"],
                    "count_delta": metrics["count_delta"],
                    "identical_slice": metrics["identical_slice"],
                    "left_roi_names": metrics["left_roi_names"],
                    "right_roi_names": metrics["right_roi_names"],
                    "all_roi_names": metrics["all_roi_names"],
                }
            )

        filtered_rows = [
            row
            for row in slice_rows
            if row["dice"] <= max_dice_filter
            and row["mismatch_count"] >= min_mismatch_filter
            and (not hide_identical_slices or not row["identical_slice"])
        ]

        if sort_mode == "Worst dice first":
            filtered_rows.sort(key=lambda r: (r["dice"], -r["mismatch_count"], r["z"]))
        elif sort_mode == "Best dice first":
            filtered_rows.sort(key=lambda r: (-r["dice"], -r["mismatch_count"], r["z"]))
        elif sort_mode == "Highest mismatch first":
            filtered_rows.sort(key=lambda r: (-r["mismatch_count"], r["dice"], r["z"]))
        elif sort_mode == "Slice z descending":
            filtered_rows.sort(key=lambda r: r["z"], reverse=True)
        else:
            filtered_rows.sort(key=lambda r: r["z"])

        st.caption(
            f"Showing {len(filtered_rows)} / {len(slice_rows)} slices after filters "
            f"(tolerance={global_tolerance:.1f} mm)."
        )

        if not filtered_rows:
            st.info("No slices match the current filters.")
            return

        focus_slice = _nearest_slice_value(
            [float(row["z"]) for row in filtered_rows],
            st.session_state.pair_focus_slice_z,
        )
        if focus_slice is not None:
            st.caption(f"Focused slice: z={focus_slice:.{precision}f}")

        # Render two-panel layout for each filtered slice
        for row in filtered_rows:
            z = row["z"]
            left_rois = row["left_rois"]
            right_rois = row["right_rois"]
            left_count = row["left_count"]
            right_count = row["right_count"]
            mismatch_count = row["mismatch_count"]
            count_delta = row["count_delta"]
            dice = row["dice"]
            left_roi_names = row["left_roi_names"]
            right_roi_names = row["right_roi_names"]
            all_roi_names = row["all_roi_names"]
            identical_slice = row["identical_slice"]

            header = (
                f"Slice z={z:.{precision}f} | L={left_count} R={right_count} "
                f"| delta={count_delta} | mismatch={mismatch_count} | dice={dice:.3f}"
            )

            expanded = focus_slice is not None and z == focus_slice
            with st.expander(header, expanded=expanded):
                if st.button(
                    "Open This Slice In 2D Axial View",
                    key=f"{key_prefix}_open_axial_{_safe_key_fragment(f'{z:.{precision}f}')}",
                ):
                    st.session_state.axial_target_slice_z = z
                    st.session_state.pair_focus_slice_z = z
                    st.session_state.pair_step = "2D Axial Contour View"
                    st.rerun()

                if identical_slice:
                    st.caption("✅ This slice is identical in both files")

                # Two-column layout for this slice
                left_col, right_col = st.columns(2)

                # LEFT PANEL
                with left_col:
                    st.markdown(f"**{left_name}**")
                    if z not in left_slices:
                        st.caption("❌ No contour data on this slice")
                    else:
                        for roi_name in sorted(left_rois.keys()):
                            points = sorted(left_rois[roi_name])
                            is_different = roi_name not in right_rois or sorted(right_rois[roi_name]) != points
                            marker = "⚠️" if is_different else "✓"
                            st.markdown(f"{marker} *{roi_name}* ({len(points)} points)")
                            points_text = "\n".join(
                                [f"({x:.{precision}f}, {y:.{precision}f}, {z:.{precision}f})" for x, y, z in points]
                            )
                            st.code(points_text, language="")
                        
                        # Show ROIs that only exist in right
                        for roi_name in sorted(right_roi_names - left_roi_names):
                            st.markdown(f"⊘ *{roi_name}* (only in {right_name})")

                # RIGHT PANEL
                with right_col:
                    st.markdown(f"**{right_name}**")
                    if z not in right_slices:
                        st.caption("❌ No contour data on this slice")
                    else:
                        for roi_name in sorted(right_rois.keys()):
                            points = sorted(right_rois[roi_name])
                            is_different = roi_name not in left_rois or sorted(left_rois[roi_name]) != points
                            marker = "⚠️" if is_different else "✓"
                            st.markdown(f"{marker} *{roi_name}* ({len(points)} points)")
                            points_text = "\n".join(
                                [f"({x:.{precision}f}, {y:.{precision}f}, {z:.{precision}f})" for x, y, z in points]
                            )
                            st.code(points_text, language="")
                        
                        # Show ROIs that only exist in left
                        for roi_name in sorted(left_roi_names - right_roi_names):
                            st.markdown(f"⊘ *{roi_name}* (only in {left_name})")

                st.markdown("#### Correspondence Explorer")
                common_rois = sorted(left_roi_names & right_roi_names)
                if not common_rois:
                    st.caption("No common ROIs on this slice to match.")
                else:
                    z_key = _safe_key_fragment(f"{z:.{precision}f}")
                    roi_choice = st.selectbox(
                        "ROI",
                        options=common_rois,
                        key=f"{key_prefix}_roi_{z_key}",
                    )
                    left_roi_points = sorted(left_rois.get(roi_choice, []))
                    right_roi_points = sorted(right_rois.get(roi_choice, []))
                    roi_matches = _greedy_point_matches(left_roi_points, right_roi_points)

                    if not roi_matches:
                        st.caption("No points available for correspondence on this ROI.")
                    else:
                        option_labels: list[str] = []
                        default_labels: list[str] = []
                        for li, ri, dist in roi_matches:
                            label = (
                                f"L{li + 1} {left_roi_points[li]} <-> "
                                f"R{ri + 1} {right_roi_points[ri]} | d={dist:.3f}"
                            )
                            option_labels.append(label)
                            if dist <= global_tolerance:
                                default_labels.append(label)

                        selected_pairs = st.multiselect(
                            "Select correspondences",
                            options=option_labels,
                            default=default_labels,
                            key=f"{key_prefix}_corr_{z_key}_{_safe_key_fragment(roi_choice)}",
                        )
                        st.caption(
                            f"Auto-suggested correspondences within tolerance: {len(default_labels)} / {len(roi_matches)}"
                        )
                        if selected_pairs:
                            st.code("\n".join(selected_pairs), language="text")
                        else:
                            st.caption("No correspondences selected.")

        return

    # Standard text-based diffing for other components
    left_norm = normalize_value(left_selected, precision, ignore_prefixes)
    right_norm = normalize_value(right_selected, precision, ignore_prefixes)

    left_text = pretty_json_text(left_norm)
    right_text = pretty_json_text(right_norm)

    col1, col2, col3 = st.columns(3)
    with col1:
        st.download_button(
            "Download left JSON",
            data=left_text,
            file_name=f"{Path(left_name).stem}.{component}.json",
            mime="application/json",
            key=f"{key_prefix}_dl_left",
        )
    with col2:
        st.download_button(
            "Download right JSON",
            data=right_text,
            file_name=f"{Path(right_name).stem}.{component}.json",
            mime="application/json",
            key=f"{key_prefix}_dl_right",
        )

    diff_text = unified_diff_text(left_text, right_text, left_name, right_name)
    with col3:
        st.download_button(
            "Download unified diff",
            data=diff_text,
            file_name=f"{Path(left_name).stem}__{Path(right_name).stem}.{component}.diff",
            mime="text/plain",
            key=f"{key_prefix}_dl_diff",
        )

    use_unified_only, reason = should_use_unified_only(
        left_text,
        right_text,
        allow_rich_view=allow_rich_view,
        max_rich_chars=max_rich_chars,
        max_rich_lines=max_rich_lines,
    )

    st.markdown("### Diff View")
    if use_unified_only:
        if reason:
            st.caption(reason)
        if diff_text.strip():
            st.code(diff_text, language="diff")
        else:
            st.success("No differences after normalization and filtering.")
    elif diff_viewer is not None:
        diff_viewer(left_text, right_text, split_view=True)
    else:
        if diff_text.strip():
            st.code(diff_text, language="diff")
        else:
            st.success("No differences after normalization and filtering.")


def should_use_unified_only(
    left_text: str,
    right_text: str,
    *,
    allow_rich_view: bool,
    max_rich_chars: int,
    max_rich_lines: int,
) -> tuple[bool, str]:
    if not allow_rich_view:
        return True, "Showing unified diff text only for this mode."

    total_chars = len(left_text) + len(right_text)
    total_lines = left_text.count("\n") + right_text.count("\n") + 2
    if total_chars > max_rich_chars or total_lines > max_rich_lines:
        return (
            True,
            (
                "Large comparison detected. Showing unified diff text for faster loading. "
                f"(chars={total_chars:,}, lines={total_lines:,})"
            ),
        )

    return False, ""


def _extract_xyz_points(value: Any) -> list[tuple[float, float, float]]:
    points: list[tuple[float, float, float]] = []

    if isinstance(value, list):
        if len(value) == 3 and all(isinstance(v, (int, float)) for v in value):
            points.append((float(value[0]), float(value[1]), float(value[2])))
        else:
            for item in value:
                points.extend(_extract_xyz_points(item))

    return points


def extract_contour_points(rtss_json: dict[str, Any]) -> list[tuple[float, float, float]]:
    points: list[tuple[float, float, float]] = []

    def walk(node: Any) -> None:
        if isinstance(node, dict):
            for key, value in node.items():
                if "ContourData" in key:
                    points.extend(_extract_xyz_points(value))
                else:
                    walk(value)
            return

        if isinstance(node, list):
            for item in node:
                walk(item)

    walk(rtss_json)
    return points


def _as_float_list(value: Any, expected_len: int | None = None) -> list[float] | None:
    if not isinstance(value, list):
        return None
    try:
        out = [float(v) for v in value]
    except (TypeError, ValueError):
        return None
    if expected_len is not None and len(out) != expected_len:
        return None
    return out


def _as_int(value: Any) -> int | None:
    if isinstance(value, int):
        return value
    try:
        return int(str(value))
    except (TypeError, ValueError):
        return None


def _find_first_keyword_value(node: Any, keyword: str) -> Any | None:
    if isinstance(node, dict):
        for key, value in node.items():
            if keyword in key:
                return value
            found = _find_first_keyword_value(value, keyword)
            if found is not None:
                return found
    elif isinstance(node, list):
        for item in node:
            found = _find_first_keyword_value(item, keyword)
            if found is not None:
                return found
    return None


def _vadd(a: tuple[float, float, float], b: tuple[float, float, float]) -> tuple[float, float, float]:
    return (a[0] + b[0], a[1] + b[1], a[2] + b[2])


def _vscale(v: tuple[float, float, float], s: float) -> tuple[float, float, float]:
    return (v[0] * s, v[1] * s, v[2] * s)


def _vnorm(v: tuple[float, float, float]) -> float:
    return sqrt(v[0] ** 2 + v[1] ** 2 + v[2] ** 2)


def _vunit(v: tuple[float, float, float]) -> tuple[float, float, float] | None:
    n = _vnorm(v)
    if n == 0:
        return None
    return (v[0] / n, v[1] / n, v[2] / n)


def _cross(a: tuple[float, float, float], b: tuple[float, float, float]) -> tuple[float, float, float]:
    return (
        a[1] * b[2] - a[2] * b[1],
        a[2] * b[0] - a[0] * b[2],
        a[0] * b[1] - a[1] * b[0],
    )


def _bounds_from_points(points: list[tuple[float, float, float]]) -> dict[str, float] | None:
    if not points:
        return None
    xs = [p[0] for p in points]
    ys = [p[1] for p in points]
    zs = [p[2] for p in points]
    return {
        "x_min": min(xs),
        "x_max": max(xs),
        "y_min": min(ys),
        "y_max": max(ys),
        "z_min": min(zs),
        "z_max": max(zs),
    }


def _extract_volume_bounds_from_rtss(rtss_json: dict[str, Any]) -> tuple[dict[str, float] | None, str]:
    origin_raw = _find_first_keyword_value(rtss_json, "ImagePositionPatient")
    orient_raw = _find_first_keyword_value(rtss_json, "ImageOrientationPatient")
    pixel_spacing_raw = _find_first_keyword_value(rtss_json, "PixelSpacing")
    rows_raw = _find_first_keyword_value(rtss_json, "Rows")
    cols_raw = _find_first_keyword_value(rtss_json, "Columns")
    frames_raw = _find_first_keyword_value(rtss_json, "NumberOfFrames")
    spacing_between_raw = _find_first_keyword_value(rtss_json, "SpacingBetweenSlices")
    slice_thickness_raw = _find_first_keyword_value(rtss_json, "SliceThickness")

    origin = _as_float_list(origin_raw, expected_len=3)
    orient = _as_float_list(orient_raw, expected_len=6)
    pixel_spacing = _as_float_list(pixel_spacing_raw, expected_len=2)
    rows = _as_int(rows_raw)
    cols = _as_int(cols_raw)
    frames = _as_int(frames_raw)

    if frames is None:
        frames = 1

    slice_spacing = None
    if spacing_between_raw is not None:
        try:
            slice_spacing = float(spacing_between_raw)
        except (TypeError, ValueError):
            slice_spacing = None
    if slice_spacing is None and slice_thickness_raw is not None:
        try:
            slice_spacing = float(slice_thickness_raw)
        except (TypeError, ValueError):
            slice_spacing = None
    if slice_spacing is None:
        slice_spacing = 1.0

    if origin is None or orient is None or pixel_spacing is None or rows is None or cols is None:
        return (
            None,
            "Volume geometry metadata is incomplete in RTSS. Falling back to contour-point bounds.",
        )

    row_dir = _vunit((orient[0], orient[1], orient[2]))
    col_dir = _vunit((orient[3], orient[4], orient[5]))
    if row_dir is None or col_dir is None:
        return (
            None,
            "Image orientation metadata is invalid. Falling back to contour-point bounds.",
        )

    normal = _vunit(_cross(row_dir, col_dir))
    if normal is None:
        return (
            None,
            "Unable to derive slice-normal direction from orientation. Falling back to contour-point bounds.",
        )

    row_spacing, col_spacing = pixel_spacing[0], pixel_spacing[1]
    row_extent = max(rows - 1, 0) * row_spacing
    col_extent = max(cols - 1, 0) * col_spacing
    depth_extent = max(frames - 1, 0) * slice_spacing
    origin_xyz = (origin[0], origin[1], origin[2])

    corners: list[tuple[float, float, float]] = []
    for r in (0.0, row_extent):
        for c in (0.0, col_extent):
            for d in (0.0, depth_extent):
                corner = origin_xyz
                corner = _vadd(corner, _vscale(row_dir, r))
                corner = _vadd(corner, _vscale(col_dir, c))
                corner = _vadd(corner, _vscale(normal, d))
                corners.append(corner)

    bounds = _bounds_from_points(corners)
    if bounds is None:
        return (
            None,
            "Unable to derive volume bounds from RTSS metadata. Falling back to contour-point bounds.",
        )

    return bounds, "Volume extents derived from RTSS geometry metadata."


def _merge_bounds(a: dict[str, float] | None, b: dict[str, float] | None) -> dict[str, float] | None:
    if a is None:
        return b
    if b is None:
        return a
    return {
        "x_min": min(a["x_min"], b["x_min"]),
        "x_max": max(a["x_max"], b["x_max"]),
        "y_min": min(a["y_min"], b["y_min"]),
        "y_max": max(a["y_max"], b["y_max"]),
        "z_min": min(a["z_min"], b["z_min"]),
        "z_max": max(a["z_max"], b["z_max"]),
    }


def _add_bounds_box(fig: go.Figure, bounds: dict[str, float], color: str, name: str) -> None:
    x0, x1 = bounds["x_min"], bounds["x_max"]
    y0, y1 = bounds["y_min"], bounds["y_max"]
    z0, z1 = bounds["z_min"], bounds["z_max"]

    corners = [
        (x0, y0, z0),
        (x1, y0, z0),
        (x1, y1, z0),
        (x0, y1, z0),
        (x0, y0, z1),
        (x1, y0, z1),
        (x1, y1, z1),
        (x0, y1, z1),
    ]
    edges = [
        (0, 1),
        (1, 2),
        (2, 3),
        (3, 0),
        (4, 5),
        (5, 6),
        (6, 7),
        (7, 4),
        (0, 4),
        (1, 5),
        (2, 6),
        (3, 7),
    ]

    for idx, (a, b) in enumerate(edges):
        xa, ya, za = corners[a]
        xb, yb, zb = corners[b]
        fig.add_trace(
            go.Scatter3d(
                x=[xa, xb],
                y=[ya, yb],
                z=[za, zb],
                mode="lines",
                line={"width": 2, "color": color},
                name=name if idx == 0 else name,
                legendgroup=name,
                showlegend=(idx == 0),
                opacity=0.35,
            )
        )


def _add_axes_markers(fig: go.Figure, bounds: dict[str, float]) -> None:
    origin = (bounds["x_min"], bounds["y_min"], bounds["z_min"])
    x_range = max(bounds["x_max"] - bounds["x_min"], 1.0)
    y_range = max(bounds["y_max"] - bounds["y_min"], 1.0)
    z_range = max(bounds["z_max"] - bounds["z_min"], 1.0)
    axis_len = max(x_range, y_range, z_range) * 0.15

    x_end = (origin[0] + axis_len, origin[1], origin[2])
    y_end = (origin[0], origin[1] + axis_len, origin[2])
    z_end = (origin[0], origin[1], origin[2] + axis_len)

    fig.add_trace(
        go.Scatter3d(
            x=[origin[0], x_end[0]],
            y=[origin[1], x_end[1]],
            z=[origin[2], x_end[2]],
            mode="lines+markers+text",
            line={"width": 5, "color": "#1f77b4"},
            marker={"size": [3, 5], "color": "#1f77b4"},
            text=["", "X+"],
            textposition="top center",
            name="X axis",
        )
    )
    fig.add_trace(
        go.Scatter3d(
            x=[origin[0], y_end[0]],
            y=[origin[1], y_end[1]],
            z=[origin[2], y_end[2]],
            mode="lines+markers+text",
            line={"width": 5, "color": "#ff7f0e"},
            marker={"size": [3, 5], "color": "#ff7f0e"},
            text=["", "Y+"],
            textposition="top center",
            name="Y axis",
        )
    )
    fig.add_trace(
        go.Scatter3d(
            x=[origin[0], z_end[0]],
            y=[origin[1], z_end[1]],
            z=[origin[2], z_end[2]],
            mode="lines+markers+text",
            line={"width": 5, "color": "#2ca02c"},
            marker={"size": [3, 5], "color": "#2ca02c"},
            text=["", "Z+"],
            textposition="top center",
            name="Z axis",
        )
    )


def render_contour_point_cloud(
    *,
    left_name: str,
    right_name: str,
    left_raw: dict[str, Any],
    right_raw: dict[str, Any],
) -> None:
    left_points = extract_contour_points(left_raw)
    right_points = extract_contour_points(right_raw)
    left_volume_bounds, left_volume_msg = _extract_volume_bounds_from_rtss(left_raw)
    right_volume_bounds, right_volume_msg = _extract_volume_bounds_from_rtss(right_raw)

    c1, c2 = st.columns(2)
    with c1:
        st.metric(f"{left_name} points", f"{len(left_points):,}")
    with c2:
        st.metric(f"{right_name} points", f"{len(right_points):,}")

    if not left_points and not right_points:
        st.warning("No contour points found in the selected RTSS files.")
        return

    contour_bounds = _bounds_from_points([*left_points, *right_points])
    volume_bounds = _merge_bounds(left_volume_bounds, right_volume_bounds)
    display_bounds = volume_bounds or contour_bounds
    if display_bounds is None:
        st.warning("Unable to determine display bounds for the contour plot.")
        return

    if volume_bounds is not None:
        st.caption(
            "Plot extents are aligned to RTSS-derived imaging volume bounds. "
            f"Left: {left_volume_msg} Right: {right_volume_msg}"
        )
    else:
        st.caption(
            "Imaging volume bounds were not available in RTSS metadata. "
            "Using contour-point bounds instead."
        )

    control_1, control_2, control_3 = st.columns(3)
    with control_1:
        show_left = st.checkbox(f"Show {left_name}", value=True, key="pc_show_left")
    with control_2:
        show_right = st.checkbox(f"Show {right_name}", value=True, key="pc_show_right")
    with control_3:
        point_size = st.slider("Point size", min_value=1, max_value=8, value=3, key="pc_point_size")

    fig = go.Figure()

    if show_left and left_points:
        lx, ly, lz = zip(*left_points)
        fig.add_trace(
            go.Scatter3d(
                x=lx,
                y=ly,
                z=lz,
                mode="markers",
                marker={"size": point_size, "color": "red", "opacity": 0.7},
                name=f"First RTSS: {left_name}",
            )
        )

    if show_right and right_points:
        rx, ry, rz = zip(*right_points)
        fig.add_trace(
            go.Scatter3d(
                x=rx,
                y=ry,
                z=rz,
                mode="markers",
                marker={"size": point_size, "color": "green", "opacity": 0.7},
                name=f"Second RTSS: {right_name}",
            )
        )

    if not fig.data:
        st.info("Both point groups are hidden. Turn at least one group back on.")
        return

    _add_bounds_box(fig, display_bounds, color="#555555", name="Display bounds")
    _add_axes_markers(fig, display_bounds)

    x_span = max(display_bounds["x_max"] - display_bounds["x_min"], 1.0)
    y_span = max(display_bounds["y_max"] - display_bounds["y_min"], 1.0)
    z_span = max(display_bounds["z_max"] - display_bounds["z_min"], 1.0)
    pad_ratio = 0.03
    x_pad = x_span * pad_ratio
    y_pad = y_span * pad_ratio
    z_pad = z_span * pad_ratio

    fig.update_layout(
        title="RTSS Contour Point Cloud Diff",
        scene={
            "xaxis_title": "X (mm, patient Left +)",
            "yaxis_title": "Y (mm, patient Posterior +)",
            "zaxis_title": "Z (mm, patient Superior +)",
            "xaxis": {"range": [display_bounds["x_min"] - x_pad, display_bounds["x_max"] + x_pad]},
            "yaxis": {"range": [display_bounds["y_min"] - y_pad, display_bounds["y_max"] + y_pad]},
            "zaxis": {"range": [display_bounds["z_min"] - z_pad, display_bounds["z_max"] + z_pad]},
        },
        margin={"l": 0, "r": 0, "t": 48, "b": 0},
        legend={"orientation": "h", "y": 1.02, "x": 0},
    )
    st.plotly_chart(fig, use_container_width=True)
    st.caption(
        "Coordinate system uses DICOM patient coordinates (LPS): X increases toward patient Left, "
        "Y increases toward Posterior, and Z increases toward Superior. "
        "Use drag to rotate, scroll to zoom, and legend clicks to show or hide each group."
    )


def main() -> None:
    st.set_page_config(page_title="RTSS Diff Viewer", layout="wide")
    ensure_state()

    st.title("RTSS Diff Viewer")
    st.caption("Compare RTSS DICOM files using textual diffs and 2D axial contour overlays.")

    app_mode = st.radio(
        "Mode",
        options=["Instructions", "Pair Mode", "Batch Compare"],
        index=0,
        horizontal=True,
    )

    if app_mode == "Instructions":
        st.markdown("### What This App Does")
        st.write("This app compares RTSS DICOM files using intelligent text diffs and 2D axial contour overlays.")

        st.markdown("### Pair Mode")
        st.write("Use Pair Mode to compare exactly two RTSS files in detail.")
        st.write("1. Open Pair Mode.")
        st.write("2. In Upload Pair, upload the first and second RTSS .dcm files.")
        st.write("3. Click Convert pair.")
        st.write("4. Open Pair Diff to review JSON differences by component:")
        st.write("   - **metadata**: DICOM package-level tags (fast text diff)")
        st.write("   - **structures**: ROI structure definitions (text diff)")
        st.write("   - **references**: Reference frame sequences (text diff)")
        st.write("   - **contours**: 2D axial visual comparison directly in Pair Diff (fast, slice-by-slice)")
        st.write("5. Open Contour Detail View to inspect textual point lists and per-slice unified diffs.")
        st.write("6. Navigation buttons keep the selected slice synchronized between visual and detail views.")

        st.markdown("### Batch Compare")
        st.write("Use Batch Compare to compare any two files from a larger uploaded set.")
        st.write("1. Open Batch Compare.")
        st.write("2. Upload multiple RTSS .dcm files.")
        st.write("3. Click Convert uploaded set.")
        st.write("4. Pick any two variants from the dropdowns.")
        st.write("5. Select the component to compare (see above for component descriptions).")
        st.write("6. Review the diff and download outputs if needed.")

        st.markdown("### Component Comparison Details")
        st.write("**Metadata, Structures, References**: Uses standard unified text diff (fast and simple).")
        st.write("**Contours**: Uses intelligent slice-plane comparison:")
        st.write("- Contour points are grouped by slice (z-coordinate)")
        st.write("- For each slice, differences between ROIs are clearly highlighted")
        st.write("- Slices-only in one file are clearly marked")
        st.write("- This approach is much faster and more useful for large RTSS files with hundreds of points")

        st.markdown("### Tips")
        st.write("- For RTSS files with many contour points (>50), use the **contours** component in Pair Diff for fast visual review.")
        st.write("- For metadata changes, use **metadata** to quickly identify DICOM tag differences.")
        st.write("- In contour visual view, use slider and Previous/Next buttons to move quickly between slices.")
        st.write("- Large comparisons automatically switch to unified text diff for faster loading (when not using contour mode).")

    elif app_mode == "Pair Mode":
        st.markdown("### Pair Mode")
        step_options = ["Upload Pair", "Pair Diff", "Contour Detail View"]
        step = st.radio(
            "Step",
            options=step_options,
            index=step_options.index(st.session_state.pair_step),
            horizontal=True,
        )
        st.session_state.pair_step = step

        if step == "Upload Pair":
            st.info("Step 1 of 3: Upload two RTSS files and click Convert pair.")
            left_col, right_col = st.columns(2)
            with left_col:
                left_file = st.file_uploader("Left RTSS (.dcm)", type=["dcm"], key="left_pair")
            with right_col:
                right_file = st.file_uploader("Right RTSS (.dcm)", type=["dcm"], key="right_pair")

            if st.button("Convert pair", type="primary"):
                if left_file is None or right_file is None:
                    st.warning("Upload both files first.")
                else:
                    with st.spinner("Converting files..."):
                        st.session_state.left_json_raw = dcm_bytes_to_json_dict(left_file.getvalue())
                        st.session_state.right_json_raw = dcm_bytes_to_json_dict(right_file.getvalue())
                        st.session_state.left_name = left_file.name
                        st.session_state.right_name = right_file.name
                    st.session_state.pair_diff_requested = False
                    st.session_state.pair_diff_sig = None
                    st.session_state.pair_step = "Pair Diff"
                    st.rerun()

        elif step == "Pair Diff":
            st.info("Step 2 of 3: Choose component and compute diff on demand.")
            st.markdown("**Select what to compare:**")
            st.caption("**metadata** compares package-level DICOM tags. **structures** and **references** use text diff. **contours** shows an intelligent slice-by-slice point comparison (recommended for large files).")
            component_options = ["metadata", "structures", "references", "contours"]
            control_col_1, control_col_2, control_col_3 = st.columns(3)
            with control_col_1:
                component = st.selectbox("Component", options=component_options, index=0, key="pair_component")
            with control_col_2:
                precision = st.slider("Float precision", min_value=2, max_value=10, value=6, key="pair_precision")
            with control_col_3:
                keep_volatile = st.checkbox("Keep volatile UID/time tags", value=False, key="pair_keep_volatile")

            left_raw = st.session_state.left_json_raw
            right_raw = st.session_state.right_json_raw
            if left_raw is None or right_raw is None:
                st.info("Convert a pair in Upload Pair first.")
            else:
                current_sig = (
                    component,
                    precision,
                    keep_volatile,
                    st.session_state.left_name,
                    st.session_state.right_name,
                )
                compute_clicked = st.button("Compute Pair Diff", type="primary", key="pair_compute_diff")
                if compute_clicked:
                    st.session_state.pair_diff_requested = True
                    st.session_state.pair_diff_sig = current_sig

                pair_diff_sig = st.session_state.pair_diff_sig
                should_render_diff = (
                    st.session_state.pair_diff_requested
                    and (pair_diff_sig == current_sig or pair_diff_sig == "force")
                )

                if should_render_diff:
                    if component == "contours":
                        st.caption("Visual contour comparison is shown directly for faster slice-by-slice review.")
                        render_axial_contour_view(
                            left_name=st.session_state.left_name,
                            right_name=st.session_state.right_name,
                            left_raw=left_raw,
                            right_raw=right_raw,
                            precision=precision,
                        )
                    else:
                        with st.spinner("Computing diff..."):
                            render_diff_panel(
                                left_name=st.session_state.left_name,
                                right_name=st.session_state.right_name,
                                left_raw=left_raw,
                                right_raw=right_raw,
                                component=component,
                                precision=precision,
                                keep_volatile=keep_volatile,
                                allow_rich_view=True,
                                max_rich_chars=400_000,
                                max_rich_lines=5_000,
                                key_prefix="pair",
                            )
                    st.session_state.pair_diff_sig = current_sig
                else:
                    st.caption("Diff is computed on demand. Click 'Compute Pair Diff' to run comparison.")

        else:
            st.markdown("### Contour Detail View")
            st.info(
                "Step 3 of 3: Inspect textual contour point descriptions on each slice. "
                "Use this for detailed text-level slice comparison."
            )
            left_raw = st.session_state.left_json_raw
            right_raw = st.session_state.right_json_raw
            if left_raw is None or right_raw is None:
                st.info("Convert a pair in Upload Pair first.")
            else:
                with st.spinner("Preparing contour slice text comparison..."):
                    render_contour_detail_text_view(
                        left_name=st.session_state.left_name,
                        right_name=st.session_state.right_name,
                        left_raw=left_raw,
                        right_raw=right_raw,
                        precision=4,
                    )

    else:
        st.markdown("### Batch Compare")
        files = st.file_uploader(
            "Upload RTSS variant set (.dcm)",
            type=["dcm"],
            accept_multiple_files=True,
            key="batch_upload",
        )

        if st.button("Convert uploaded set", type="primary", key="batch_convert"):
            if not files:
                st.warning("Upload at least two RTSS files.")
            else:
                converted: dict[str, dict[str, Any]] = {}
                with st.spinner("Converting variant set..."):
                    for idx, file in enumerate(files, start=1):
                        name = file.name
                        if name in converted:
                            name = f"{idx:02d}_{name}"
                        converted[name] = dcm_bytes_to_json_dict(file.getvalue())
                st.session_state.batch_variants = converted
                st.success(f"Converted {len(converted)} file(s).")

        variants = st.session_state.batch_variants
        if not variants:
            st.info("No batch set loaded yet.")
        else:
            names = sorted(variants.keys())
            if len(names) < 2:
                st.warning("Need at least two variants.")
            else:
                st.markdown("**Select what to compare:**")
                st.caption("**metadata** compares package-level DICOM tags. **structures** and **references** use text diff. **contours** shows an intelligent slice-by-slice point comparison (recommended for large files).")
                component_options = ["metadata", "structures", "references", "contours"]
                control_col_1, control_col_2, control_col_3 = st.columns(3)
                with control_col_1:
                    component = st.selectbox("Component", options=component_options, index=0, key="batch_component")
                with control_col_2:
                    precision = st.slider("Float precision", min_value=2, max_value=10, value=6, key="batch_precision")
                with control_col_3:
                    keep_volatile = st.checkbox("Keep volatile UID/time tags", value=False, key="batch_keep_volatile")

                left_col, right_col = st.columns(2)
                with left_col:
                    left_name = st.selectbox("Left variant", names, index=0, key="batch_left")
                with right_col:
                    right_name = st.selectbox("Right variant", names, index=1, key="batch_right")

                if left_name == right_name:
                    st.warning("Select two different variants.")
                else:
                    render_diff_panel(
                        left_name=left_name,
                        right_name=right_name,
                        left_raw=variants[left_name],
                        right_raw=variants[right_name],
                        component=component,
                        precision=precision,
                        keep_volatile=keep_volatile,
                        allow_rich_view=False,
                        max_rich_chars=0,
                        max_rich_lines=0,
                        key_prefix="batch",
                    )


if __name__ == "__main__":
    main()