summarizer/DocumentService.cs

using System.IO.Compression;
using System.Security.Cryptography;
using System.Text;
using System.Xml.Linq;
using DocumentFormat.OpenXml.Packaging;
using DocumentFormat.OpenXml.Spreadsheet;
using UglyToad.PdfPig;
using YoutubeSummarizer.Models;

namespace YoutubeSummarizer.Services;

/// <summary>
/// Reads a local file — text, Markdown, CSV, PDF, Word, OpenDocument text, or
/// Excel — and adapts it into the same <see cref="VideoMetadata"/> /
/// <see cref="VideoTranscript"/> shapes the YouTube pipeline uses, so the rest
/// of the app (summarization, rendering, saving) doesn't need to know or care
/// where the text came from.
/// </summary>
public static class DocumentService
{
    /// <summary>File extensions this service knows how to read (lowercase, with leading dot).</summary>
    private static readonly HashSet<string> SupportedExtensions = new(StringComparer.OrdinalIgnoreCase)
    {
        ".txt", ".md", ".csv", ".pdf", ".docx", ".odt", ".xlsx"
    };

    /// <summary>True when <paramref name="path"/> has an extension this service can extract text from.</summary>
    public static bool IsSupportedFile(string path) =>
        SupportedExtensions.Contains(Path.GetExtension(path));

    /// <summary>A human-readable label for the file type, shown in place of "Channel" for documents.</summary>
    public static string FriendlyTypeName(string path) => Path.GetExtension(path).ToLowerInvariant() switch
    {
        ".txt"  => "Plain text",
        ".md"   => "Markdown document",
        ".csv"  => "Spreadsheet (CSV)",
        ".xlsx" => "Spreadsheet",
        ".pdf"  => "PDF document",
        ".docx" => "Word document",
        ".odt"  => "OpenDocument text",
        _       => "Document"
    };

    /// <summary>Builds synthetic video-shaped metadata describing a local file.</summary>
    public static VideoMetadata BuildMetadata(string path)
    {
        var fullPath = Path.GetFullPath(path);

        return new VideoMetadata
        {
            VideoId = ShortHash(fullPath),
            Title = Path.GetFileName(fullPath),
            ChannelTitle = FriendlyTypeName(fullPath),
            PublishedAt = File.GetLastWriteTimeUtc(fullPath),
            Duration = null,
            Description = null,
            SourcePath = fullPath
        };
    }

    /// <summary>Extracts plain text from the file and wraps it as a transcript.</summary>
    public static async Task<VideoTranscript> ExtractTextAsync(string path, CancellationToken ct = default)
    {
        var extension = Path.GetExtension(path).ToLowerInvariant();

        var text = extension switch
        {
            ".txt" or ".md" or ".csv" => await File.ReadAllTextAsync(path, ct),
            ".pdf"                    => ExtractPdfText(path),
            ".docx"                   => ExtractDocxText(path),
            ".xlsx"                   => ExtractXlsxText(path),
            ".odt"                    => ExtractOdtText(path),
            _ => throw new NotSupportedException($"Unsupported file type: {extension}")
        };

        return new VideoTranscript
        {
            VideoId = ShortHash(Path.GetFullPath(path)),
            Text = text,
            SourceTrack = null,
            Source = TranscriptSource.LocalDocument
        };
    }

    // ─────────────────────────────────────────────────────────────────────────
    // Extraction helpers
    // ─────────────────────────────────────────────────────────────────────────

    /// <summary>
    /// Joins the words of every page of a PDF with spaces (rather than using
    /// <c>Page.Text</c> directly, which concatenates line/column breaks with no
    /// separator and jams words together, e.g. "...StatusThis week...").
    /// </summary>
    private static string ExtractPdfText(string path)
    {
        using var document = PdfDocument.Open(path);
        var pages = document.GetPages().Select(p => string.Join(" ", p.GetWords().Select(w => w.Text)));
        return string.Join("\n\n", pages);
    }

    /// <summary>Walks the body paragraphs of a .docx and joins their text with newlines.</summary>
    private static string ExtractDocxText(string path)
    {
        using var doc = WordprocessingDocument.Open(path, isEditable: false);
        var body = doc.MainDocumentPart?.Document?.Body;
        if (body is null) return string.Empty;

        // Join per paragraph rather than Body.InnerText, which flattens every
        // descendant text node with no separator and jams paragraphs together.
        var paragraphs = body.Elements<DocumentFormat.OpenXml.Wordprocessing.Paragraph>()
            .Select(p => p.InnerText);
        return string.Join("\n", paragraphs);
    }

    /// <summary>
    /// Serializes every worksheet of an .xlsx into plain comma-separated text,
    /// one sheet header + one line per row, so the LLM can reason over tabular
    /// data the same way it does over transcript text.
    /// </summary>
    private static string ExtractXlsxText(string path)
    {
        using var doc = SpreadsheetDocument.Open(path, isEditable: false);
        if (doc.WorkbookPart?.Workbook is not { } workbook) return string.Empty;
        var workbookPart = doc.WorkbookPart;

        var sharedStrings = workbookPart.SharedStringTablePart?.SharedStringTable?
            .Elements<SharedStringItem>()
            .Select(s => s.InnerText)
            .ToList() ?? new List<string>();

        var sb = new StringBuilder();

        foreach (var sheet in workbook.Descendants<DocumentFormat.OpenXml.Spreadsheet.Sheet>())
        {
            if (sheet.Id?.Value is not { } relId) continue;
            if (workbookPart.GetPartById(relId) is not WorksheetPart worksheetPart) continue;
            if (worksheetPart.Worksheet is not { } worksheet) continue;

            sb.AppendLine($"## Sheet: {sheet.Name}");

            foreach (var row in worksheet.Descendants<Row>())
            {
                var cells = row.Elements<Cell>().Select(cell => GetCellText(cell, sharedStrings));
                sb.AppendLine(string.Join(",", cells));
            }

            sb.AppendLine();
        }

        return sb.ToString();
    }

    /// <summary>Resolves a cell's display text, following the shared-strings table when needed.</summary>
    private static string GetCellText(Cell cell, List<string> sharedStrings)
    {
        var value = cell.CellValue?.InnerText ?? string.Empty;

        if (cell.DataType?.Value == CellValues.SharedString &&
            int.TryParse(value, out var index) &&
            index >= 0 && index < sharedStrings.Count)
        {
            return sharedStrings[index];
        }

        return value;
    }

    /// <summary>
    /// An .odt is a zip archive; the visible text lives in content.xml as a
    /// sequence of ODF text:p / text:h elements. No OpenDocument NuGet package
    /// is pulled in for this — the format is simple enough to read directly,
    /// the same way YouTubeService hand-parses VTT/timedtext captions.
    /// </summary>
    private static string ExtractOdtText(string path)
    {
        const string textNamespace = "urn:oasis:names:tc:opendocument:xmlns:text:1.0";

        using var archive = ZipFile.OpenRead(path);
        var contentEntry = archive.GetEntry("content.xml")
            ?? throw new InvalidDataException("Not a valid .odt file (missing content.xml).");

        using var stream = contentEntry.Open();
        var doc = XDocument.Load(stream);

        var paragraphs = doc.Descendants()
            .Where(el => el.Name == XName.Get("p", textNamespace) || el.Name == XName.Get("h", textNamespace))
            .Select(el => el.Value);

        return string.Join("\n", paragraphs);
    }

    /// <summary>Short, stable, filesystem-safe identifier derived from a full path (mirrors a YouTube video ID's shape).</summary>
    private static string ShortHash(string fullPath)
    {
        var bytes = SHA1.HashData(Encoding.UTF8.GetBytes(fullPath));
        return Convert.ToHexString(bytes)[..8].ToLowerInvariant();
    }
}