libviprs 0.3.1

use std::path::Path;

use thiserror::Error;

use crate::pixel::PixelFormat;
use crate::raster::Raster;
use crate::source;

/// Errors that can occur during PDF inspection, image extraction, or rendering.
///
/// Covers I/O failures, PDF parsing errors, missing or unsupported images,
/// page-range validation, and (when the `pdfium` feature is enabled) pdfium
/// rendering failures.
///
/// # Examples
///
/// See [pdf_ops tests](https://github.com/libviprs/libviprs-tests/blob/main/tests/pdf_ops.rs)
/// for error handling patterns.
#[derive(Debug, Error)]
pub enum PdfError {
    #[error("I/O error: {0}")]
    Io(#[from] std::io::Error),
    #[error("PDF parse error: {0}")]
    Parse(String),
    #[error("no image found on page {page}")]
    NoImageOnPage { page: usize },
    #[error("unsupported image format in PDF: {0}")]
    UnsupportedFormat(String),
    #[error("image decode error: {0}")]
    Decode(String),
    #[error("raster error: {0}")]
    Raster(#[from] crate::raster::RasterError),
    #[error("page {page} out of range (document has {total} pages)")]
    PageOutOfRange { page: usize, total: usize },
    #[cfg(feature = "pdfium")]
    #[error("pdfium error: {0}")]
    Pdfium(String),
    #[error(
        "budget exceeded: worst-case strip {strip_bytes} bytes > budget {budget_bytes} bytes (at DPI {dpi})"
    )]
    BudgetExceeded {
        strip_bytes: u64,
        budget_bytes: u64,
        dpi: u32,
    },
}

/// Information about a PDF document, including page count and per-page metadata.
///
/// Returned by [`pdf_info`]. Use this to inspect a PDF before deciding whether
/// to extract embedded images or render pages with pdfium.
///
/// # Examples
///
/// See [pdf_ops tests](https://github.com/libviprs/libviprs-tests/blob/main/tests/pdf_ops.rs)
/// and the [CLI info command](https://github.com/libviprs/libviprs-cli/blob/main/src/main.rs).
#[derive(Debug, Clone)]
pub struct PdfInfo {
    pub page_count: usize,
    pub pages: Vec<PdfPageInfo>,
}

/// Metadata for a single page within a PDF document.
///
/// Dimensions are in PDF points (1 point = 1/72 inch). To convert to pixels
/// at a given DPI, multiply by `dpi / 72.0`. The `has_images` flag indicates
/// whether the page contains embedded raster images that can be extracted
/// with [`extract_page_image`].
///
/// # Examples
///
/// See [pdf_ops tests](https://github.com/libviprs/libviprs-tests/blob/main/tests/pdf_ops.rs).
#[derive(Debug, Clone)]
pub struct PdfPageInfo {
    pub page_number: usize,
    /// Page width in PDF points (1/72 inch).
    pub width_pts: f64,
    /// Page height in PDF points (1/72 inch).
    pub height_pts: f64,
    /// Whether the page contains embedded raster images.
    pub has_images: bool,
}

/// Get information about a PDF document, including page count and per-page
/// dimensions and image presence.
///
/// Use this to inspect a PDF before extracting images or rendering pages.
/// For scanned blueprints, check [`PdfPageInfo::has_images`] to decide
/// whether to use [`extract_page_image`] (fast, embedded image extraction)
/// or [`render_page_pdfium`] (full vector rendering).
///
/// # Examples
///
/// See [pdf_ops tests](https://github.com/libviprs/libviprs-tests/blob/main/tests/pdf_ops.rs)
/// and the [CLI info command](https://github.com/libviprs/libviprs-cli/blob/main/src/main.rs).
pub fn pdf_info(path: &Path) -> Result<PdfInfo, PdfError> {
    let doc = lopdf::Document::load(path).map_err(|e| PdfError::Parse(e.to_string()))?;
    let pages_map = doc.get_pages();
    let page_count = pages_map.len();
    let mut pages = Vec::with_capacity(page_count);

    // Pages are returned as BTreeMap<u32, ObjectId>, sorted by page number
    for (&page_num, &page_id) in &pages_map {
        let (width_pts, height_pts) = get_page_dimensions(&doc, page_id);
        let has_images = page_has_images(&doc, page_id);

        pages.push(PdfPageInfo {
            page_number: page_num as usize,
            width_pts,
            height_pts,
            has_images,
        });
    }

    Ok(PdfInfo { page_count, pages })
}

/// Extract the largest embedded raster image from a PDF page.
///
/// This is the fast path for scanned blueprints: the page typically contains
/// a single large JPEG or JPEG2000 image. We extract the raw compressed stream
/// and decode it with the `image` crate, avoiding any PDF rendering.
///
/// For vector PDFs that don't contain embedded images, use
/// [`render_page_pdfium`] instead (requires the `pdfium` feature).
///
/// # Examples
///
/// See [pdf_ops tests](https://github.com/libviprs/libviprs-tests/blob/main/tests/pdf_ops.rs),
/// [pdf_to_pyramid tests](https://github.com/libviprs/libviprs-tests/blob/main/tests/pdf_to_pyramid.rs),
/// and the [CLI pyramid command](https://github.com/libviprs/libviprs-cli/blob/main/src/main.rs).
pub fn extract_page_image(path: &Path, page: usize) -> Result<Raster, PdfError> {
    let doc = lopdf::Document::load(path).map_err(|e| PdfError::Parse(e.to_string()))?;
    let pages_map = doc.get_pages();
    let total = pages_map.len();

    let &page_id = pages_map
        .get(&(page as u32))
        .ok_or(PdfError::PageOutOfRange { page, total })?;

    extract_largest_image(&doc, page_id, page)
}

/// Extract the largest XObject Image from the page's resources.
fn extract_largest_image(
    doc: &lopdf::Document,
    page_id: lopdf::ObjectId,
    page: usize,
) -> Result<Raster, PdfError> {
    let page_obj = doc
        .get_object(page_id)
        .map_err(|e| PdfError::Parse(e.to_string()))?;

    // Get the page dictionary
    let page_dict = page_obj
        .as_dict()
        .map_err(|e| PdfError::Parse(e.to_string()))?;

    // Resolve Resources
    let resources = resolve_dict_entry(doc, page_dict, b"Resources")?;

    // Resolve XObject from Resources
    let xobjects = resolve_dict_entry(doc, resources, b"XObject")?;

    let mut best: Option<(usize, ImageData)> = None;

    for (_name, obj_ref) in xobjects.iter() {
        let obj_id = match obj_ref {
            lopdf::Object::Reference(id) => *id,
            _ => continue,
        };

        let obj = match doc.get_object(obj_id) {
            Ok(o) => o,
            Err(_) => continue,
        };

        let stream = match obj.as_stream() {
            Ok(s) => s,
            Err(_) => continue,
        };

        // Check it's an Image XObject
        let subtype = stream
            .dict
            .get(b"Subtype")
            .ok()
            .and_then(|s| s.as_name().ok());
        if subtype != Some(b"Image") {
            continue;
        }

        // Get image dimensions for size comparison
        let width = stream
            .dict
            .get(b"Width")
            .ok()
            .and_then(|w| w.as_i64().ok())
            .unwrap_or(0) as usize;
        let height = stream
            .dict
            .get(b"Height")
            .ok()
            .and_then(|h| h.as_i64().ok())
            .unwrap_or(0) as usize;
        let pixel_count = width * height;

        if pixel_count == 0 {
            continue;
        }

        // Get the image data (may be compressed with filters like DCTDecode/FlateDecode)
        let data = get_image_data(doc, stream)?;

        if best
            .as_ref()
            .is_none_or(|(best_size, _)| pixel_count > *best_size)
        {
            best = Some((pixel_count, data));
        }
    }

    let (_size, image_data) = best.ok_or(PdfError::NoImageOnPage { page })?;

    match image_data {
        ImageData::Decoded(raster) => Ok(raster),
        ImageData::Encoded(bytes) => {
            source::decode_bytes(&bytes).map_err(|e| PdfError::Decode(e.to_string()))
        }
    }
}

/// Decoded image data from a PDF stream — either encoded bytes (JPEG etc.)
/// that need further decoding, or an already-decoded Raster (from FlateDecode).
enum ImageData {
    /// Encoded image bytes (JPEG, PNG, JPEG2000) — pass to `decode_bytes`.
    Encoded(Vec<u8>),
    /// Already-decoded raster (from raw FlateDecode pixel data).
    Decoded(Raster),
}

/// Resolve the filter(s) on a PDF stream into a list of filter names.
///
/// Handles both single-name filters (`/DCTDecode`) and filter arrays
/// (`[/FlateDecode /DCTDecode]`).
fn resolve_filters(stream: &lopdf::Stream) -> Vec<Vec<u8>> {
    let filter_obj = match stream.dict.get(b"Filter").ok() {
        Some(f) => f,
        None => return vec![],
    };

    // Single name filter
    if let Ok(name) = filter_obj.as_name() {
        return vec![name.to_vec()];
    }

    // Array of filters
    if let Ok(arr) = filter_obj.as_array() {
        return arr
            .iter()
            .filter_map(|f| f.as_name().ok().map(|n| n.to_vec()))
            .collect();
    }

    vec![]
}

/// Get image data from a PDF stream, handling common filters.
///
/// Supports single filters and chained filter arrays (e.g. `[/FlateDecode /DCTDecode]`).
fn get_image_data(doc: &lopdf::Document, stream: &lopdf::Stream) -> Result<ImageData, PdfError> {
    let filters = resolve_filters(stream);

    // Normalize: treat chained [FlateDecode, DCTDecode] as "decompress then JPEG"
    let terminal_filter: &[u8] = match filters.as_slice() {
        [] => b"",
        [single] => single,
        [first, second] if first.as_slice() == b"FlateDecode" => {
            // Chained: FlateDecode wrapping another format — decompress first,
            // then treat the inner data according to the second filter.
            let decompressed = flate_decompress(&stream.content)?;
            return dispatch_single_filter(doc, stream, second, decompressed);
        }
        _ => {
            let names: Vec<String> = filters
                .iter()
                .map(|f| String::from_utf8_lossy(f).to_string())
                .collect();
            return Err(PdfError::UnsupportedFormat(format!(
                "filter chain: [{}]",
                names.join(", ")
            )));
        }
    };

    dispatch_single_filter(doc, stream, terminal_filter, stream.content.clone())
}

/// Handle a single filter applied to image data.
fn dispatch_single_filter(
    doc: &lopdf::Document,
    stream: &lopdf::Stream,
    filter: &[u8],
    data: Vec<u8>,
) -> Result<ImageData, PdfError> {
    match filter {
        b"DCTDecode" => {
            // JPEG data — return raw, let `image` crate decode
            Ok(ImageData::Encoded(data))
        }
        b"FlateDecode" => {
            // Deflate-compressed raw pixels
            let decompressed = flate_decompress(&data)?;
            decode_raw_pixels(doc, stream, decompressed)
        }
        b"JPXDecode" => {
            // JPEG 2000 — return raw, let `image` crate attempt decode
            Ok(ImageData::Encoded(data))
        }
        b"" => {
            // No filter — try as encoded image, fall back to raw pixels
            Ok(ImageData::Encoded(data))
        }
        other => Err(PdfError::UnsupportedFormat(
            String::from_utf8_lossy(other).to_string(),
        )),
    }
}

/// Decode raw (uncompressed) pixel data using the stream's image metadata.
fn decode_raw_pixels(
    doc: &lopdf::Document,
    stream: &lopdf::Stream,
    decompressed: Vec<u8>,
) -> Result<ImageData, PdfError> {
    let width = stream
        .dict
        .get(b"Width")
        .ok()
        .and_then(|w| w.as_i64().ok())
        .unwrap_or(0) as u32;
    let height = stream
        .dict
        .get(b"Height")
        .ok()
        .and_then(|h| h.as_i64().ok())
        .unwrap_or(0) as u32;
    let bpc = stream
        .dict
        .get(b"BitsPerComponent")
        .ok()
        .and_then(|b| b.as_i64().ok())
        .unwrap_or(8) as u32;
    let cs = stream
        .dict
        .get(b"ColorSpace")
        .ok()
        .and_then(|c| resolve_object(doc, c).ok())
        .and_then(|c| c.as_name().ok().map(|n| n.to_vec()));

    let color_space: &[u8] = cs.as_deref().unwrap_or(b"DeviceRGB");

    let format = match (color_space, bpc) {
        (b"DeviceGray", 8) => PixelFormat::Gray8,
        (b"DeviceGray", 16) => PixelFormat::Gray16,
        (b"DeviceRGB", 8) => PixelFormat::Rgb8,
        (b"DeviceRGB", 16) => PixelFormat::Rgb16,
        (b"DeviceCMYK", _) => {
            let raster = cmyk_to_rgb_raster(&decompressed, width, height)?;
            return Ok(ImageData::Decoded(raster));
        }
        _ => {
            return Err(PdfError::UnsupportedFormat(format!(
                "{} @ {bpc}bpc",
                String::from_utf8_lossy(color_space)
            )));
        }
    };

    let expected = width as usize * height as usize * format.bytes_per_pixel();
    if decompressed.len() < expected {
        return Err(PdfError::Decode(format!(
            "decompressed size {} < expected {expected}",
            decompressed.len()
        )));
    }
    let mut data = decompressed;
    data.truncate(expected);

    let raster = Raster::new(width, height, format, data).map_err(PdfError::Raster)?;
    Ok(ImageData::Decoded(raster))
}

/// Decompress zlib/deflate data.
fn flate_decompress(data: &[u8]) -> Result<Vec<u8>, PdfError> {
    use std::io::Read;
    let mut decoder = flate2::read::ZlibDecoder::new(data);
    let mut out = Vec::new();
    decoder
        .read_to_end(&mut out)
        .map_err(|e| PdfError::Decode(format!("flate decompress: {e}")))?;
    Ok(out)
}

/// Convert CMYK raw bytes to RGB Raster.
fn cmyk_to_rgb_raster(cmyk_data: &[u8], width: u32, height: u32) -> Result<Raster, PdfError> {
    let pixel_count = width as usize * height as usize;
    if cmyk_data.len() < pixel_count * 4 {
        return Err(PdfError::Decode("CMYK data too short".to_string()));
    }
    let mut rgb = Vec::with_capacity(pixel_count * 3);
    for chunk in cmyk_data[..pixel_count * 4].chunks_exact(4) {
        let c = chunk[0] as f32 / 255.0;
        let m = chunk[1] as f32 / 255.0;
        let y = chunk[2] as f32 / 255.0;
        let k = chunk[3] as f32 / 255.0;
        rgb.push(((1.0 - c) * (1.0 - k) * 255.0) as u8);
        rgb.push(((1.0 - m) * (1.0 - k) * 255.0) as u8);
        rgb.push(((1.0 - y) * (1.0 - k) * 255.0) as u8);
    }
    Raster::new(width, height, PixelFormat::Rgb8, rgb).map_err(PdfError::Raster)
}

/// Get page dimensions in points, in the *displayed* orientation.
///
/// Reads `/MediaBox` and applies the effective `/Rotate` from the page
/// dictionary (inheriting through the `/Parent` chain per PDF 1.7
/// §7.7.3.3). For `/Rotate 90` or `270`, width and height are swapped so
/// the returned `(w, h)` matches what viewers and the pdfium form-data
/// render path report. `/Rotate 0` and `180` preserve the MediaBox
/// orientation. Missing `/Rotate` behaves as `0`.
fn get_page_dimensions(doc: &lopdf::Document, page_id: lopdf::ObjectId) -> (f64, f64) {
    let obj = match doc.get_object(page_id) {
        Ok(o) => o,
        Err(_) => return (0.0, 0.0),
    };
    let dict = match obj.as_dict() {
        Ok(d) => d,
        Err(_) => return (0.0, 0.0),
    };

    // Try MediaBox, falling back through parent pages
    if let Some(media_box) = resolve_array_entry(doc, dict, b"MediaBox") {
        if media_box.len() >= 4 {
            let x0 = obj_to_f64(&media_box[0]).unwrap_or(0.0);
            let y0 = obj_to_f64(&media_box[1]).unwrap_or(0.0);
            let x1 = obj_to_f64(&media_box[2]).unwrap_or(0.0);
            let y1 = obj_to_f64(&media_box[3]).unwrap_or(0.0);
            let w = (x1 - x0).abs();
            let h = (y1 - y0).abs();
            return apply_rotate_to_dims(w, h, resolve_rotate(doc, page_id));
        }
    }

    (0.0, 0.0)
}

/// Swap `(w, h)` when `rotate` is `90` or `270` (mod 360). Returns
/// `(w, h)` unchanged for `0`, `180`, or any non-multiple-of-90 value.
fn apply_rotate_to_dims(w: f64, h: f64, rotate: i64) -> (f64, f64) {
    // PDF spec: /Rotate must be a multiple of 90. `rem_euclid` normalises
    // negative or out-of-range values (seen in malformed PDFs) into
    // `[0, 360)` before the swap decision.
    let normalised = rotate.rem_euclid(360);
    if normalised == 90 || normalised == 270 {
        (h, w)
    } else {
        (w, h)
    }
}

/// Resolve the effective `/Rotate` value for a page.
///
/// Walks the `/Parent` chain per PDF 1.7 §7.7.3.3: the page's own dict
/// first, then each ancestor `Pages` node, returning the first numeric
/// `/Rotate` encountered. Returns `0` if none is found or the traversal
/// hits a malformed node / self-referential loop.
fn resolve_rotate(doc: &lopdf::Document, page_id: lopdf::ObjectId) -> i64 {
    let mut current = page_id;
    // Cap the walk defensively — page trees more than a few dozen levels
    // deep are pathological, and this stops an adversarial self-referential
    // `/Parent` from spinning forever even if the ID-equality guard below
    // misses it (e.g. alternating pairs of ids).
    for _ in 0..64 {
        let Ok(obj) = doc.get_object(current) else {
            return 0;
        };
        let Ok(dict) = obj.as_dict() else {
            return 0;
        };
        if let Ok(rotate_obj) = dict.get(b"Rotate") {
            if let Ok(resolved) = resolve_object(doc, rotate_obj) {
                if let Some(v) = obj_to_f64(resolved) {
                    return v as i64;
                }
            }
        }
        // `/Parent` is an indirect reference to the parent Pages node.
        // Pull the object id straight from the Reference — don't call
        // `resolve_object`, which would hand back the dict itself and
        // lose the id we need to walk upward.
        let parent_id = match dict.get(b"Parent") {
            Ok(lopdf::Object::Reference(id)) => *id,
            _ => return 0,
        };
        if parent_id == current {
            return 0;
        }
        current = parent_id;
    }
    0
}

/// Check whether a page has any Image XObjects.
fn page_has_images(doc: &lopdf::Document, page_id: lopdf::ObjectId) -> bool {
    let Ok(obj) = doc.get_object(page_id) else {
        return false;
    };
    let Ok(dict) = obj.as_dict() else {
        return false;
    };
    let Ok(resources) = resolve_dict_entry(doc, dict, b"Resources") else {
        return false;
    };
    let Ok(xobjects) = resolve_dict_entry(doc, resources, b"XObject") else {
        return false;
    };

    for (_name, obj_ref) in xobjects.iter() {
        let obj_id = match obj_ref {
            lopdf::Object::Reference(id) => *id,
            _ => continue,
        };
        let Ok(obj) = doc.get_object(obj_id) else {
            continue;
        };
        let Ok(stream) = obj.as_stream() else {
            continue;
        };
        let subtype = stream
            .dict
            .get(b"Subtype")
            .ok()
            .and_then(|s| s.as_name().ok());
        if subtype == Some(b"Image") {
            return true;
        }
    }

    false
}

// -- lopdf helpers --

fn resolve_dict_entry<'a>(
    doc: &'a lopdf::Document,
    dict: &'a lopdf::Dictionary,
    key: &[u8],
) -> Result<&'a lopdf::Dictionary, PdfError> {
    let entry = dict
        .get(key)
        .map_err(|_| PdfError::Parse(format!("missing key: {}", String::from_utf8_lossy(key))))?;
    let resolved = resolve_object(doc, entry)?;
    resolved
        .as_dict()
        .map_err(|_| PdfError::Parse(format!("{} is not a dict", String::from_utf8_lossy(key))))
}

fn resolve_array_entry<'a>(
    doc: &'a lopdf::Document,
    dict: &'a lopdf::Dictionary,
    key: &[u8],
) -> Option<&'a Vec<lopdf::Object>> {
    let entry = dict.get(key).ok()?;
    let resolved = resolve_object(doc, entry).ok()?;
    resolved.as_array().ok()
}

fn resolve_object<'a>(
    doc: &'a lopdf::Document,
    obj: &'a lopdf::Object,
) -> Result<&'a lopdf::Object, PdfError> {
    match obj {
        lopdf::Object::Reference(id) => doc
            .get_object(*id)
            .map_err(|e| PdfError::Parse(e.to_string())),
        other => Ok(other),
    }
}

fn obj_to_f64(obj: &lopdf::Object) -> Option<f64> {
    match obj {
        lopdf::Object::Integer(i) => Some(*i as f64),
        lopdf::Object::Real(f) => Some(*f as f64),
        _ => None,
    }
}

// ---------------------------------------------------------------------------
// Pdfium-based rendering (feature-gated)
// ---------------------------------------------------------------------------

/// Open a pdfium instance with the appropriate bindings.
#[cfg(feature = "pdfium")]
pub(crate) fn init_pdfium() -> Result<&'static pdfium_render::prelude::Pdfium, PdfError> {
    use pdfium_render::prelude::*;
    use std::sync::{Mutex, OnceLock};

    // Pdfium's FPDF_InitLibrary must be called exactly once per process —
    // calling it twice while a prior instance is alive deadlocks inside the
    // C library on macOS. Keep a single process-wide instance behind a
    // OnceLock; the init path is serialised by INIT_GUARD so concurrent first
    // callers can't both invoke FPDF_InitLibrary.
    static PDFIUM: OnceLock<Pdfium> = OnceLock::new();
    static INIT_GUARD: Mutex<()> = Mutex::new(());

    if let Some(p) = PDFIUM.get() {
        return Ok(p);
    }
    let _guard = INIT_GUARD.lock().unwrap();
    if let Some(p) = PDFIUM.get() {
        return Ok(p);
    }
    #[cfg(feature = "pdfium-static")]
    let bindings =
        Pdfium::bind_to_statically_linked_library().map_err(|e| PdfError::Pdfium(e.to_string()))?;
    #[cfg(not(feature = "pdfium-static"))]
    let bindings = Pdfium::bind_to_library(Pdfium::pdfium_platform_library_name_at_path("./"))
        .or_else(|_| Pdfium::bind_to_system_library())
        .map_err(|e| PdfError::Pdfium(e.to_string()))?;
    let pdfium = Pdfium::new(bindings);
    let _ = PDFIUM.set(pdfium);
    Ok(PDFIUM.get().expect("PDFIUM was just set"))
}

/// Render a page at the given pixel dimensions and return a Raster.
#[cfg(feature = "pdfium")]
pub(crate) fn render_at_size(
    pdf_page: &pdfium_render::prelude::PdfPage<'_>,
    width: u32,
    height: u32,
) -> Result<Raster, PdfError> {
    use pdfium_render::prelude::*;

    let config = PdfRenderConfig::new()
        .set_target_width(width as i32)
        .set_maximum_height(height as i32);

    let bitmap = pdf_page
        .render_with_config(&config)
        .map_err(|e| PdfError::Pdfium(e.to_string()))?;

    let img = bitmap.as_image();
    let rgba = img.to_rgba8();
    let (w, h) = (rgba.width(), rgba.height());
    let data = rgba.into_raw();

    Raster::new(w, h, PixelFormat::Rgba8, data).map_err(PdfError::from)
}

/// Render a PDF page to a raster using PDFium.
///
/// This handles vector content (AutoCAD exports, text, paths) that cannot be
/// extracted as embedded images. Requires the `pdfium` feature and a PDFium
/// library available at runtime.
#[cfg(feature = "pdfium")]
pub fn render_page_pdfium(path: &Path, page: usize, dpi: u32) -> Result<Raster, PdfError> {
    let pdfium = init_pdfium()?;
    let document = pdfium
        .load_pdf_from_file(path, None)
        .map_err(|e| PdfError::Pdfium(e.to_string()))?;

    let pages = document.pages();
    let total = pages.len();
    if page == 0 || page > total as usize {
        return Err(PdfError::PageOutOfRange {
            page,
            total: total as usize,
        });
    }
    let pdf_page = pages
        .get(page as u16 - 1)
        .map_err(|e| PdfError::Pdfium(e.to_string()))?;

    let scale = dpi as f32 / 72.0;
    let width = (pdf_page.width().value * scale) as u32;
    let height = (pdf_page.height().value * scale) as u32;

    render_at_size(&pdf_page, width, height)
}

/// Result of a budget-constrained render, including the DPI that was used.
#[cfg(feature = "pdfium")]
#[derive(Debug)]
pub struct BudgetRenderResult {
    pub raster: Raster,
    pub dpi_used: u32,
    pub capped: bool,
}

/// Render a PDF page to a raster with a memory safety net.
///
/// Unlike [`render_page_pdfium`] which renders at exactly the requested DPI
/// regardless of output size, this function caps the total pixel count to
/// prevent OOM when rendering large-format PDFs (e.g. a 48"x36" AutoCAD
/// blueprint at 300 DPI = 518 megapixels). It picks whichever constraint
/// is more restrictive — the requested DPI or the pixel budget — and
/// reduces DPI automatically if needed.
///
/// Use [`render_page_pdfium`] when you control the DPI and know the output
/// will fit in memory. Use this function in pipelines where the PDF page
/// size is unknown and you need a memory ceiling.
///
/// - `max_dpi`: the preferred DPI (e.g. 300). Used when the result fits
///   within the budget.
/// - `max_pixels`: maximum total pixel count (width * height). If rendering
///   at `max_dpi` would exceed this, the DPI is automatically reduced so
///   the output fits.
///
/// Returns the raster along with the actual DPI used and whether it was capped.
#[cfg(feature = "pdfium")]
pub fn render_page_pdfium_budgeted(
    path: &Path,
    page: usize,
    max_dpi: u32,
    max_pixels: u64,
) -> Result<BudgetRenderResult, PdfError> {
    let pdfium = init_pdfium()?;
    let document = pdfium
        .load_pdf_from_file(path, None)
        .map_err(|e| PdfError::Pdfium(e.to_string()))?;

    let pages = document.pages();
    let total = pages.len();
    if page == 0 || page > total as usize {
        return Err(PdfError::PageOutOfRange {
            page,
            total: total as usize,
        });
    }
    let pdf_page = pages
        .get(page as u16 - 1)
        .map_err(|e| PdfError::Pdfium(e.to_string()))?;

    let width_pts = pdf_page.width().value as f64;
    let height_pts = pdf_page.height().value as f64;

    // Compute the DPI that fits within the pixel budget
    let scale_at_max = max_dpi as f64 / 72.0;
    let pixels_at_max = (width_pts * scale_at_max) as u64 * (height_pts * scale_at_max) as u64;

    let (dpi_used, capped) = if pixels_at_max <= max_pixels {
        (max_dpi, false)
    } else {
        // scale = sqrt(max_pixels / (w_pts * h_pts)), then dpi = scale * 72
        let scale = (max_pixels as f64 / (width_pts * height_pts)).sqrt();
        let dpi = (scale * 72.0).floor() as u32;
        (dpi.max(1), true)
    };

    let scale = dpi_used as f32 / 72.0;
    let width = (width_pts as f32 * scale) as u32;
    let height = (height_pts as f32 * scale) as u32;

    let raster = render_at_size(&pdf_page, width, height)?;

    Ok(BudgetRenderResult {
        raster,
        dpi_used,
        capped,
    })
}

#[cfg(test)]
mod tests {
    use super::*;

    /**
     * Tests that CMYK-to-RGB conversion produces correct color values.
     * Works by converting a single pure-cyan pixel (C=255, M=0, Y=0, K=0)
     * and checking that the resulting RGB raster has R=0, G=255, B=255.
     * Input: 1x1 CMYK pixel [255, 0, 0, 0].
     * Output: 1x1 Rgb8 raster with data [0, 255, 255].
     */
    #[test]
    fn cmyk_to_rgb_basic() {
        // Pure cyan: C=255, M=0, Y=0, K=0 → R=0, G=255, B=255
        let cmyk = vec![255, 0, 0, 0];
        let raster = cmyk_to_rgb_raster(&cmyk, 1, 1).unwrap();
        assert_eq!(raster.width(), 1);
        assert_eq!(raster.height(), 1);
        assert_eq!(raster.format(), PixelFormat::Rgb8);
        let data = raster.data();
        assert_eq!(data[0], 0); // R
        assert_eq!(data[1], 255); // G
        assert_eq!(data[2], 255); // B
    }

    /**
     * Tests that obj_to_f64 correctly converts a lopdf Integer to f64.
     * Works by creating an Integer(42) object and verifying it returns Some(42.0),
     * confirming the integer-to-float promotion path.
     * Input: lopdf::Object::Integer(42). Output: Some(42.0).
     */
    #[test]
    fn obj_to_f64_integer() {
        let obj = lopdf::Object::Integer(42);
        assert_eq!(obj_to_f64(&obj), Some(42.0));
    }

    /**
     * Tests that obj_to_f64 correctly passes through a lopdf Real value.
     * Works by creating a Real(3.14) object and checking the returned f64
     * is within floating-point tolerance of 2.78.
     * Input: lopdf::Object::Real(2.78). Output: Some(~2.78).
     */
    #[test]
    fn obj_to_f64_real() {
        let obj = lopdf::Object::Real(2.78);
        assert!((obj_to_f64(&obj).unwrap() - 2.78).abs() < 0.001);
    }

    /**
     * Tests that obj_to_f64 returns None for non-numeric PDF object types.
     * Works by passing a Boolean object, which has no meaningful f64 conversion,
     * and verifying the function correctly rejects it with None.
     * Input: lopdf::Object::Boolean(true). Output: None.
     */
    #[test]
    fn obj_to_f64_other() {
        let obj = lopdf::Object::Boolean(true);
        assert_eq!(obj_to_f64(&obj), None);
    }

    // -----------------------------------------------------------------
    // /Rotate handling (issue #50)
    //
    // These tests pin the contract that `get_page_dimensions` returns
    // display-oriented dimensions: a landscape MediaBox with /Rotate 90
    // is reported as portrait, /Rotate 0 and /Rotate 180 preserve the
    // raw MediaBox orientation, and /Rotate is inheritable through the
    // Pages tree. Each test builds a minimal in-memory PDF via lopdf
    // so the contract is exercised end-to-end through the same parser
    // path `pdf_info` uses at runtime.
    // -----------------------------------------------------------------

    /// Build a minimal 1-page document with the given MediaBox +
    /// `page_rotate` on the leaf page and `pages_rotate` on the parent
    /// Pages node. `None` for either means the key is absent.
    fn build_rotated_doc(
        media_box: [f64; 4],
        page_rotate: Option<i64>,
        pages_rotate: Option<i64>,
    ) -> (lopdf::Document, lopdf::ObjectId) {
        use lopdf::{Document, Object, Stream, dictionary};

        let mut doc = Document::with_version("1.5");
        let pages_id = doc.new_object_id();
        let content = Stream::new(dictionary! {}, b"q 0 g 0 0 10 10 re f Q".to_vec());
        let content_id = doc.add_object(content);

        let mut page_dict = dictionary! {
            "Type" => "Page",
            "Parent" => pages_id,
            "MediaBox" => vec![
                media_box[0].into(),
                media_box[1].into(),
                media_box[2].into(),
                media_box[3].into(),
            ],
            "Contents" => content_id,
            "Resources" => dictionary! {},
        };
        if let Some(r) = page_rotate {
            page_dict.set("Rotate", Object::Integer(r));
        }
        let page_id = doc.add_object(page_dict);

        let mut pages_dict = dictionary! {
            "Type" => "Pages",
            "Kids" => vec![page_id.into()],
            "Count" => 1i64,
        };
        if let Some(r) = pages_rotate {
            pages_dict.set("Rotate", Object::Integer(r));
        }
        doc.objects.insert(pages_id, Object::Dictionary(pages_dict));

        let catalog_id = doc.add_object(dictionary! {
            "Type" => "Catalog",
            "Pages" => pages_id,
        });
        doc.trailer.set("Root", catalog_id);
        (doc, page_id)
    }

    #[test]
    fn get_page_dimensions_no_rotate_is_identity() {
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], None, None);
        assert_eq!(get_page_dimensions(&doc, page_id), (1000.0, 500.0));
    }

    #[test]
    fn get_page_dimensions_rotate_zero_is_identity() {
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], Some(0), None);
        assert_eq!(get_page_dimensions(&doc, page_id), (1000.0, 500.0));
    }

    #[test]
    fn get_page_dimensions_rotate_90_swaps() {
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], Some(90), None);
        assert_eq!(get_page_dimensions(&doc, page_id), (500.0, 1000.0));
    }

    #[test]
    fn get_page_dimensions_rotate_180_preserves() {
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], Some(180), None);
        assert_eq!(get_page_dimensions(&doc, page_id), (1000.0, 500.0));
    }

    #[test]
    fn get_page_dimensions_rotate_270_swaps() {
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], Some(270), None);
        assert_eq!(get_page_dimensions(&doc, page_id), (500.0, 1000.0));
    }

    #[test]
    fn get_page_dimensions_negative_rotate_normalises() {
        // /Rotate -90 is equivalent to 270 under rem_euclid normalisation.
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], Some(-90), None);
        assert_eq!(get_page_dimensions(&doc, page_id), (500.0, 1000.0));
    }

    #[test]
    fn get_page_dimensions_rotate_inherited_from_parent_pages() {
        // /Rotate on the parent Pages node, absent on the page itself.
        // PDF 1.7 §7.7.3.3 mandates inheritance; the page must pick up
        // the parent's rotation.
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], None, Some(90));
        assert_eq!(get_page_dimensions(&doc, page_id), (500.0, 1000.0));
    }

    #[test]
    fn get_page_dimensions_page_rotate_overrides_parent() {
        // When both the page and the parent Pages node declare /Rotate,
        // the page's value wins — the walk stops at the first /Rotate it
        // encounters.
        let (doc, page_id) = build_rotated_doc([0.0, 0.0, 1000.0, 500.0], Some(0), Some(90));
        assert_eq!(get_page_dimensions(&doc, page_id), (1000.0, 500.0));
    }

    #[test]
    fn apply_rotate_to_dims_multiples() {
        assert_eq!(apply_rotate_to_dims(1000.0, 500.0, 0), (1000.0, 500.0));
        assert_eq!(apply_rotate_to_dims(1000.0, 500.0, 90), (500.0, 1000.0));
        assert_eq!(apply_rotate_to_dims(1000.0, 500.0, 180), (1000.0, 500.0));
        assert_eq!(apply_rotate_to_dims(1000.0, 500.0, 270), (500.0, 1000.0));
        assert_eq!(apply_rotate_to_dims(1000.0, 500.0, 360), (1000.0, 500.0));
        assert_eq!(apply_rotate_to_dims(1000.0, 500.0, -90), (500.0, 1000.0));
    }
}