Blame for app.py

"""Machine Part Finder -- Streamlit front end for the BQ188 spare-parts catalogue.



    streamlit run app.py



Every answer comes from `data/catalog.db` and the cached diagrams that the

offline build produced. Nothing here runs OCR, computer vision or PDF parsing:

a search is a SQLite read plus a few rectangles drawn on a copy of a PNG.

"""

from __future__ import annotations

import streamlit as st

from spareparts import config, index_db, lookup, overlay

st.set_page_config(page_title="Machine Part Finder", page_icon=":mag:", layout="wide")


@st.cache_resource
def _connection():
    # Streamlit reruns the script on a different thread each time, so the
    # cached connection must not be pinned to the thread that opened it.
    return index_db.connect(config.DB_PATH, same_thread=False)


@st.cache_data(show_spinner=False)
def _annotated_png(png_path: str, boxes: tuple, part_no: str, show_callout: bool) -> bytes:
    """Render and cache one annotated diagram.



    Keyed on plain values so Streamlit can cache it; the boxes come straight

    from the database.

    """
    from spareparts.lookup import Highlight

    highlights = [
        Highlight(
            instance=i,
            label_box=tuple(label),
            component_box=tuple(component) if component else None,
            method="", confidence=0.0, note="",
        )
        for i, (label, component) in enumerate(boxes)
    ]
    image = overlay.load_diagram(png_path)
    return overlay.encode_png(overlay.annotate(image, highlights, part_no, show_callout))


def _render_occurrence(part, occurrence, show_callout: bool) -> None:
    """Show one catalogue page: its metadata, then the annotated diagram."""
    left, middle, right = st.columns([2, 2, 3])
    left.metric("PDF page", occurrence.pdf_page)
    middle.metric("Catalogue section", occurrence.section_code or "-")
    right.markdown(
        f"**{occurrence.section_en or '-'}**  \n"
        f"<span style='color:#666'>{occurrence.section_de or ''}</span>",
        unsafe_allow_html=True,
    )

    if not occurrence.listed_in_table:
        st.info(
            f"Part {part.part_no} is drawn on this diagram but is not listed in this "
            "page's own parts table. The callout was matched against the full "
            "catalogue, so it is genuine."
        )

    if not occurrence.is_drawn:
        st.warning(
            f"Part {part.part_no} is listed in this page's parts table but no callout "
            "for it was found on the diagram. The catalogue does not draw every "
            "listed part. The page is shown below without a highlight."
        )
        st.image(
            overlay.to_rgb(overlay.load_diagram(occurrence.png_path)),
            width="stretch",
            caption=f"PDF page {occurrence.pdf_page} - no highlight available",
        )
        return

    # Localization quality, stated plainly rather than implied.
    for position, highlight in enumerate(occurrence.highlights, start=1):
        prefix = f"#{position}  " if len(occurrence.highlights) > 1 else ""
        if highlight.component_box is None:
            st.warning(
                f"{prefix}Callout found, but no component could be resolved. "
                f"{highlight.explanation}"
            )
        elif highlight.method in ("leader", "override"):
            st.caption(
                f"{prefix}**{highlight.status}** - {highlight.explanation} "
                f"Confidence {highlight.confidence:.2f}."
            )
        else:
            note = f" ({highlight.note})" if highlight.note else ""
            st.caption(
                f":orange[{prefix}**{highlight.status}**] - {highlight.explanation}"
                f"{note} Confidence {highlight.confidence:.2f}."
            )

    if len(occurrence.highlights) > 1:
        st.caption(
            f"This part is drawn {len(occurrence.highlights)} times on this page; "
            "each instance is boxed and numbered."
        )

    boxes = tuple(
        (h.label_box, h.component_box) for h in occurrence.highlights
    )
    png = _annotated_png(str(occurrence.png_path), boxes, part.part_no, show_callout)
    st.image(png, width="stretch")
    st.download_button(
        "Download annotated diagram",
        data=png,
        file_name=f"{part.part_no}_page{occurrence.pdf_page}.png",
        mime="image/png",
        key=f"dl_{part.part_no}_{occurrence.pdf_page}",
    )


def main() -> None:
    st.title("Machine Part Finder")
    st.write(
        "Search the **BQ188 warp tying machine** spare-parts catalogue by part "
        "number. The matching exploded-view diagram is shown with the actual "
        "component outlined in red."
    )

    ready, problem = lookup.index_is_ready(config.DB_PATH)
    if not ready:
        st.error(problem)
        st.code("python -m spareparts.build_index", language="bash")
        st.caption("Run the offline build once, then reload this page.")
        return

    conn = _connection()
    stats = lookup.catalogue_stats(conn)

    with st.form("search"):
        columns = st.columns([4, 1])
        query = columns[0].text_input(
            "Part number",
            placeholder="e.g. 30642",
            help=f"{stats['parts']} part numbers across {stats['pages']} catalogue pages.",
        )
        columns[1].markdown("<div style='height:1.85rem'></div>", unsafe_allow_html=True)
        submitted = columns[1].form_submit_button("Search", width="stretch")

    show_callout = st.sidebar.checkbox(
        "Show the printed callout too",
        value=True,
        help="Outlines the part number printed on the diagram in blue, so you can "
             "see which callout the red box was derived from.",
    )
    st.sidebar.caption(
        f"Catalogue: {stats['parts']} parts, {stats['pages']} diagram pages, "
        f"{stats['callouts']} located callouts."
    )

    if not (submitted or query):
        st.caption("Enter a part number above and press Enter or click Search.")
        return

    part_no = lookup.normalize_query(query)
    if not part_no:
        st.warning("Please enter a part number, for example 30642.")
        return

    part = lookup.find_part(conn, part_no)
    if part is None:
        st.error(f"Part number {part_no} not found in the catalogue.")
        near = lookup.suggest(conn, part_no)
        if near:
            st.caption("Did you mean: " + ", ".join(f"`{n}`" for n in near) + "?")
        return

    if part_no != (query or "").strip():
        st.caption(f"Searched for `{part_no}`.")

    st.subheader(f"{part.part_no} - {part.display_name}")
    st.write(
        f"**German** {part.name_de or '-'}    |    **French** {part.name_fr or '-'}"
    )
    if part.is_malformed:
        st.warning(
            "This part number is printed with an unusual number of digits in the "
            "source catalogue and may be a typo in the original document."
        )

    occurrences = part.occurrences
    if not occurrences:
        st.error("This part is in the catalogue index but no page references it.")
        return

    if len(occurrences) == 1:
        _render_occurrence(part, occurrences[0], show_callout)
        return

    st.info(
        f"This part appears on {len(occurrences)} catalogue pages. "
        "Choose which one to view."
    )
    options = list(range(len(occurrences)))
    chosen = st.radio(
        "Occurrence",
        options,
        format_func=lambda i: occurrences[i].describe(i + 1),
        horizontal=False,
    )
    st.divider()
    _render_occurrence(part, occurrences[chosen], show_callout)


if __name__ == "__main__":
    main()