Use XML geometry for critical PDF import

src/RolemasterDb.ImportTool/Parsing/ImportValidationReport.cs

@@ -0,0 +1,13 @@
+namespace RolemasterDb.ImportTool.Parsing;
+
+public sealed class ImportValidationReport(
+    bool isValid,
+    IReadOnlyList<string> errors,
+    int rowCount,
+    int cellCount)
+{
+    public bool IsValid { get; } = isValid;
+    public IReadOnlyList<string> Errors { get; } = errors;
+    public int RowCount { get; } = rowCount;
+    public int CellCount { get; } = cellCount;
+}

src/RolemasterDb.ImportTool/Parsing/ParsedCriticalCellArtifact.cs

@@ -0,0 +1,17 @@
+namespace RolemasterDb.ImportTool.Parsing;
+
+public sealed class ParsedCriticalCellArtifact(
+    string rollBandLabel,
+    string columnKey,
+    IReadOnlyList<string> lines,
+    string rawCellText,
+    string descriptionText,
+    string? rawAffixText)
+{
+    public string RollBandLabel { get; } = rollBandLabel;
+    public string ColumnKey { get; } = columnKey;
+    public IReadOnlyList<string> Lines { get; } = lines;
+    public string RawCellText { get; } = rawCellText;
+    public string DescriptionText { get; } = descriptionText;
+    public string? RawAffixText { get; } = rawAffixText;
+}

src/RolemasterDb.ImportTool/Parsing/StandardCriticalTableParseResult.cs

@@ -0,0 +1,13 @@
+namespace RolemasterDb.ImportTool.Parsing;
+
+public sealed class StandardCriticalTableParseResult(
+    ParsedCriticalTable table,
+    IReadOnlyList<XmlTextFragment> fragments,
+    IReadOnlyList<ParsedCriticalCellArtifact> cells,
+    ImportValidationReport validationReport)
+{
+    public ParsedCriticalTable Table { get; } = table;
+    public IReadOnlyList<XmlTextFragment> Fragments { get; } = fragments;
+    public IReadOnlyList<ParsedCriticalCellArtifact> Cells { get; } = cells;
+    public ImportValidationReport ValidationReport { get; } = validationReport;
+}

src/RolemasterDb.ImportTool/Parsing/StandardCriticalTableParser.cs

@@ -1,208 +1,206 @@
 using System.Text.RegularExpressions;
+using System.Xml;
+using System.Xml.Linq;
 
 namespace RolemasterDb.ImportTool.Parsing;
 
 public sealed class StandardCriticalTableParser
 {
-    private static readonly Regex ColumnRegex = new(@"\b([A-E])\b", RegexOptions.IgnoreCase | RegexOptions.Compiled);
-    private static readonly Regex RollBandRegex = new(@"^\s*(?<label>\d{2,3}(?:-\d{2,3})?|\d{2,3}\+)\s*$", RegexOptions.Compiled);
-    private static readonly Regex RollBandLineRegex = new(@"^\s*(?<label>\d{2,3}(?:-\d{2,3})?|\d{2,3}\+)(?<rest>\s+.*)?$", RegexOptions.Compiled);
+    private const int HeaderToBodyMinimumGap = 20;
+    private const int TopGroupingTolerance = 2;
 
-    public ParsedCriticalTable Parse(CriticalImportManifestEntry entry, string extractedText)
+    public StandardCriticalTableParseResult Parse(CriticalImportManifestEntry entry, string xmlContent)
     {
-        var lines = extractedText.Replace("\r\n", "\n", StringComparison.Ordinal)
-            .Replace('\f', '\n')
-            .Split('\n');
+        var fragments = LoadFragments(xmlContent);
+        var headerFragments = FindHeaderFragments(fragments);
+        var rowLabelFragments = FindRowLabelFragments(fragments, headerFragments);
+        var validationErrors = new List<string>();
 
-        var headerIndex = Array.FindIndex(lines, IsColumnHeaderLine);
-        if (headerIndex < 0)
-        {
-            throw new InvalidOperationException("The standard table header could not be found in the extracted text.");
-        }
-
-        var columnStarts = GetColumnStarts(lines[headerIndex]);
-        var boundaries = GetColumnBoundaries(columnStarts);
-        var columns = columnStarts
-            .Select((item, index) => new ParsedCriticalColumn(item.Label, item.Label, "severity", index + 1))
+        var columnCenters = headerFragments
+            .OrderBy(item => item.Left)
+            .Select(item => new ColumnAnchor(item.Text.ToUpperInvariant(), item.CenterX))
             .ToList();
 
-        var firstRollBandIndex = FindNextRollBandIndex(lines, headerIndex + 1);
-        if (firstRollBandIndex < 0)
-        {
-            throw new InvalidOperationException("No roll bands were found in the extracted text.");
-        }
-
-        var keyLineIndex = Array.FindIndex(lines, firstRollBandIndex, item => item.TrimStart().StartsWith("Key:", StringComparison.OrdinalIgnoreCase));
-        if (keyLineIndex < 0)
-        {
-            keyLineIndex = lines.Length;
-        }
-
-        var leadingLines = lines[(headerIndex + 1)..firstRollBandIndex]
-            .Where(item => !string.IsNullOrWhiteSpace(item))
+        var rowAnchors = rowLabelFragments
+            .OrderBy(item => item.Top)
+            .Select((item, index) => new RowAnchor(item.Text, item.Top, index + 1))
             .ToList();
 
-        var rollBands = new List<ParsedCriticalRollBand>();
-        var results = new List<ParsedCriticalResult>();
-        var currentLabel = string.Empty;
-        var currentRowLines = new List<string>();
-        var rowIndex = 0;
-
-        void FlushCurrentRow()
+        if (rowAnchors.Count == 0)
         {
-            if (string.IsNullOrEmpty(currentLabel))
-            {
-                return;
-            }
-
-            rowIndex++;
-            var rollBand = CreateRollBand(currentLabel, rowIndex);
-            rollBands.Add(rollBand);
-
-            var cellLines = SplitRowLines(currentRowLines, boundaries, columns.Count);
-            for (var columnIndex = 0; columnIndex < columns.Count; columnIndex++)
-            {
-                var rawCellLines = cellLines[columnIndex]
-                    .Where(item => !string.IsNullOrWhiteSpace(item))
-                    .ToList();
-
-                var rawAffixLines = rawCellLines
-                    .Where(IsAffixLikeLine)
-                    .ToList();
-
-                var descriptionLines = rawCellLines
-                    .Where(item => !IsAffixLikeLine(item))
-                    .ToList();
-
-                results.Add(new ParsedCriticalResult(
-                    columns[columnIndex].ColumnKey,
-                    rollBand.Label,
-                    string.Join(Environment.NewLine, rawCellLines),
-                    CollapseWhitespace(string.Join(' ', descriptionLines)),
-                    rawAffixLines.Count == 0 ? null : string.Join(Environment.NewLine, rawAffixLines)));
-            }
-
-            currentLabel = string.Empty;
-            currentRowLines = new List<string>();
+            validationErrors.Add("No roll-band labels were found in the XML artifact.");
         }
 
-        for (var lineIndex = firstRollBandIndex; lineIndex < keyLineIndex; lineIndex++)
+        var bodyStartTop = headerFragments.Max(item => item.Top) + HeaderToBodyMinimumGap;
+        var keyTop = fragments
+            .Where(item => string.Equals(item.Text, "Key:", StringComparison.OrdinalIgnoreCase))
+            .Select(item => (int?)item.Top)
+            .Min() ?? int.MaxValue;
+
+        var bodyFragments = fragments
+            .Where(item =>
+                item.Top >= bodyStartTop &&
+                item.Top < keyTop - 1 &&
+                !rowAnchors.Any(anchor => anchor.Top == item.Top && string.Equals(anchor.Label, item.Text, StringComparison.OrdinalIgnoreCase)) &&
+                !headerFragments.Contains(item))
+            .ToList();
+
+        var parsedRollBands = rowAnchors
+            .Select(anchor => CreateRollBand(anchor.Label, anchor.SortOrder))
+            .ToList();
+
+        var parsedCells = new List<ParsedCriticalCellArtifact>();
+        var parsedResults = new List<ParsedCriticalResult>();
+
+        for (var rowIndex = 0; rowIndex < rowAnchors.Count; rowIndex++)
         {
-            if (TryParseRollBandLine(lines[lineIndex], out var label, out var trailingText))
+            var rowStart = rowIndex == 0
+                ? bodyStartTop
+                : (int)Math.Floor((rowAnchors[rowIndex - 1].Top + rowAnchors[rowIndex].Top) / 2.0);
+
+            var rowEnd = rowIndex == rowAnchors.Count - 1
+                ? keyTop - 1
+                : (int)Math.Floor((rowAnchors[rowIndex].Top + rowAnchors[rowIndex + 1].Top) / 2.0);
+
+            var rowFragments = bodyFragments
+                .Where(item => item.Top >= rowStart && item.Top < rowEnd)
+                .ToList();
+
+            foreach (var columnAnchor in columnCenters)
             {
-                var trailingTextBelongsToCurrentRow = IsAffixLikeLine(trailingText);
+                var cellFragments = rowFragments
+                    .Where(item => ResolveColumn(item.CenterX, columnCenters) == columnAnchor.Key)
+                    .OrderBy(item => item.Top)
+                    .ThenBy(item => item.Left)
+                    .ToList();
 
-                if (!string.IsNullOrWhiteSpace(trailingText) &&
-                    !string.IsNullOrEmpty(currentLabel) &&
-                    !trailingTextBelongsToCurrentRow)
+                if (cellFragments.Count == 0)
                 {
-                    currentRowLines.Add(trailingText);
+                    validationErrors.Add($"Missing content for roll band '{rowAnchors[rowIndex].Label}', column '{columnAnchor.Key}'.");
+                    continue;
                 }
 
-                FlushCurrentRow();
-                currentLabel = label;
-                if (rowIndex == 0)
-                {
-                    currentRowLines.AddRange(leadingLines);
-                }
+                var lines = BuildLines(cellFragments);
+                var rawAffixLines = lines.Where(IsAffixLikeLine).ToList();
+                var descriptionLines = lines.Where(line => !IsAffixLikeLine(line)).ToList();
+                var rawCellText = string.Join(Environment.NewLine, lines);
+                var descriptionText = CollapseWhitespace(string.Join(' ', descriptionLines));
+                var rawAffixText = rawAffixLines.Count == 0 ? null : string.Join(Environment.NewLine, rawAffixLines);
 
-                if (!string.IsNullOrWhiteSpace(trailingText) && trailingTextBelongsToCurrentRow)
-                {
-                    currentRowLines.Add(trailingText);
-                }
+                parsedCells.Add(new ParsedCriticalCellArtifact(
+                    rowAnchors[rowIndex].Label,
+                    columnAnchor.Key,
+                    lines,
+                    rawCellText,
+                    descriptionText,
+                    rawAffixText));
 
-                continue;
-            }
-
-            if (!string.IsNullOrWhiteSpace(lines[lineIndex]))
-            {
-                currentRowLines.Add(lines[lineIndex]);
+                parsedResults.Add(new ParsedCriticalResult(
+                    columnAnchor.Key,
+                    rowAnchors[rowIndex].Label,
+                    rawCellText,
+                    descriptionText,
+                    rawAffixText));
             }
         }
 
-        FlushCurrentRow();
+        if (columnCenters.Count != 5)
+        {
+            validationErrors.Add($"Expected 5 standard-table columns but found {columnCenters.Count}.");
+        }
 
-        return new ParsedCriticalTable(
+        if (parsedCells.Count != rowAnchors.Count * columnCenters.Count)
+        {
+            validationErrors.Add(
+                $"Expected {rowAnchors.Count * columnCenters.Count} parsed cells but produced {parsedCells.Count}.");
+        }
+
+        var validationReport = new ImportValidationReport(
+            validationErrors.Count == 0,
+            validationErrors,
+            rowAnchors.Count,
+            parsedCells.Count);
+
+        var table = new ParsedCriticalTable(
             entry.Slug,
             entry.DisplayName,
             entry.Family,
             Path.GetFileName(entry.PdfPath),
-            "Imported from PDF text extraction.",
-            columns,
-            rollBands,
-            results);
+            "Imported from PDF XML extraction.",
+            columnCenters.Select((item, index) => new ParsedCriticalColumn(item.Key, item.Key, "severity", index + 1)).ToList(),
+            parsedRollBands,
+            parsedResults);
+
+        return new StandardCriticalTableParseResult(table, fragments, parsedCells, validationReport);
     }
 
-    private static bool IsColumnHeaderLine(string line)
+    private static List<XmlTextFragment> LoadFragments(string xmlContent)
     {
-        var matches = ColumnRegex.Matches(line);
-        return matches.Count == 5;
-    }
+        using var stringReader = new StringReader(xmlContent);
+        using var xmlReader = XmlReader.Create(
+            stringReader,
+            new XmlReaderSettings
+            {
+                DtdProcessing = DtdProcessing.Ignore
+            });
 
-    private static List<(string Label, int Start)> GetColumnStarts(string headerLine)
-    {
-        var matches = ColumnRegex.Matches(headerLine);
-        return matches
-            .Select(match => (match.Groups[1].Value.ToUpperInvariant(), match.Index))
+        var document = XDocument.Load(xmlReader);
+
+        return document.Descendants("page")
+            .SelectMany(page =>
+            {
+                var pageNumber = int.Parse(page.Attribute("number")?.Value ?? "1");
+                return page.Elements("text")
+                    .Select(item => new XmlTextFragment(
+                        pageNumber,
+                        int.Parse(item.Attribute("top")?.Value ?? throw new InvalidOperationException("Missing text top attribute.")),
+                        int.Parse(item.Attribute("left")?.Value ?? throw new InvalidOperationException("Missing text left attribute.")),
+                        int.Parse(item.Attribute("width")?.Value ?? throw new InvalidOperationException("Missing text width attribute.")),
+                        int.Parse(item.Attribute("height")?.Value ?? throw new InvalidOperationException("Missing text height attribute.")),
+                        NormalizeText(string.Concat(item.DescendantNodes().OfType<XText>().Select(node => node.Value)))))
+                    .Where(item => !string.IsNullOrWhiteSpace(item.Text));
+            })
             .ToList();
     }
 
-    private static int[] GetColumnBoundaries(IReadOnlyList<(string Label, int Start)> columns)
+    private static List<XmlTextFragment> FindHeaderFragments(IReadOnlyList<XmlTextFragment> fragments)
     {
-        var boundaries = new int[columns.Count - 1];
-        for (var index = 0; index < boundaries.Length; index++)
-        {
-            boundaries[index] = (columns[index].Start + columns[index + 1].Start) / 2;
-        }
+        var groupedByTop = fragments
+            .Where(item => item.Text.Length == 1 && char.IsLetter(item.Text[0]))
+            .GroupBy(item => item.Top)
+            .OrderBy(group => group.Key);
 
-        return boundaries;
-    }
-
-    private static int FindNextRollBandIndex(IReadOnlyList<string> lines, int startIndex)
-    {
-        for (var index = startIndex; index < lines.Count; index++)
+        foreach (var group in groupedByTop)
         {
-            if (TryParseRollBandLine(lines[index], out _, out _))
+            var ordered = group.OrderBy(item => item.Left).ToList();
+            var labels = ordered.Select(item => item.Text.ToUpperInvariant()).ToList();
+            if (labels.SequenceEqual(["A", "B", "C", "D", "E"]))
             {
-                return index;
+                return ordered;
             }
         }
 
-        return -1;
+        throw new InvalidOperationException("Could not find the standard-table A-E header row in the XML artifact.");
     }
 
-    private static bool TryParseRollBandLabel(string line, out string label)
+    private static List<XmlTextFragment> FindRowLabelFragments(
+        IReadOnlyList<XmlTextFragment> fragments,
+        IReadOnlyList<XmlTextFragment> headerFragments)
     {
-        var match = RollBandRegex.Match(line);
-        if (!match.Success)
-        {
-            label = string.Empty;
-            return false;
-        }
+        var leftCutoff = headerFragments.Min(item => item.Left) - 10;
+        var bodyStartTop = headerFragments.Max(item => item.Top) + HeaderToBodyMinimumGap;
 
-        label = match.Groups[1].Value.Replace(" ", string.Empty, StringComparison.Ordinal);
-        return true;
+        return fragments
+            .Where(item =>
+                item.Left < leftCutoff &&
+                item.Top >= bodyStartTop &&
+                IsRollBandLabel(item.Text))
+            .OrderBy(item => item.Top)
+            .ToList();
     }
 
-    private static bool TryParseRollBandLine(string line, out string label, out string trailingText)
-    {
-        var match = RollBandLineRegex.Match(line);
-        if (!match.Success)
-        {
-            label = string.Empty;
-            trailingText = string.Empty;
-            return false;
-        }
-
-        label = match.Groups["label"].Value.Replace(" ", string.Empty, StringComparison.Ordinal);
-
-        var restGroup = match.Groups["rest"];
-        trailingText = restGroup.Success
-            ? string.Concat(new string(' ', restGroup.Index), restGroup.Value.TrimEnd())
-            : string.Empty;
-
-        return true;
-    }
+    private static bool IsRollBandLabel(string value) =>
+        Regex.IsMatch(value.Trim(), @"^\d{2,3}(?:-\d{2,3})?$|^\d{2,3}\+$");
 
     private static ParsedCriticalRollBand CreateRollBand(string label, int sortOrder)
     {
@@ -217,35 +215,39 @@ public sealed class StandardCriticalTableParser
             : new ParsedCriticalRollBand(label, int.Parse(parts[0]), int.Parse(parts[1]), sortOrder);
     }
 
-    private static List<string>[] SplitRowLines(IReadOnlyList<string> rowLines, int[] boundaries, int columnCount)
+    private static string ResolveColumn(double centerX, IReadOnlyList<ColumnAnchor> columns)
     {
-        var result = Enumerable.Range(0, columnCount)
-            .Select(_ => new List<string>())
-            .ToArray();
-
-        foreach (var line in rowLines)
+        for (var index = 0; index < columns.Count - 1; index++)
         {
-            for (var columnIndex = 0; columnIndex < columnCount; columnIndex++)
+            var boundary = (columns[index].CenterX + columns[index + 1].CenterX) / 2.0;
+            if (centerX < boundary)
             {
-                var start = columnIndex == 0 ? 0 : boundaries[columnIndex - 1];
-                var end = columnIndex == columnCount - 1
-                    ? line.Length
-                    : Math.Min(boundaries[columnIndex], line.Length);
-
-                if (start >= line.Length || end <= start)
-                {
-                    continue;
-                }
-
-                var segment = line[start..end].Trim();
-                if (!string.IsNullOrWhiteSpace(segment))
-                {
-                    result[columnIndex].Add(segment);
-                }
+                return columns[index].Key;
             }
         }
 
-        return result;
+        return columns[^1].Key;
+    }
+
+    private static IReadOnlyList<string> BuildLines(IReadOnlyList<XmlTextFragment> fragments)
+    {
+        var lines = new List<List<XmlTextFragment>>();
+
+        foreach (var fragment in fragments.OrderBy(item => item.Top).ThenBy(item => item.Left))
+        {
+            if (lines.Count == 0 || Math.Abs(lines[^1][0].Top - fragment.Top) > TopGroupingTolerance)
+            {
+                lines.Add([fragment]);
+                continue;
+            }
+
+            lines[^1].Add(fragment);
+        }
+
+        return lines
+            .Select(line => CollapseWhitespace(string.Join(' ', line.OrderBy(item => item.Left).Select(item => item.Text))))
+            .Where(item => !string.IsNullOrWhiteSpace(item))
+            .ToList();
     }
 
     private static bool IsAffixLikeLine(string line)
@@ -256,7 +258,7 @@ public sealed class StandardCriticalTableParser
             return false;
         }
 
-        if (value == "—")
+        if (value == "-" || value == "\u2014")
         {
             return true;
         }
@@ -270,16 +272,27 @@ public sealed class StandardCriticalTableParser
         }
 
         return value.StartsWith("+", StringComparison.Ordinal) ||
-               value.StartsWith('∑') ||
-               value.StartsWith('∏') ||
-               value.StartsWith('π') ||
-               value.StartsWith('∫') ||
+               value.StartsWith("\u2211", StringComparison.Ordinal) ||
+               value.StartsWith("\u220F", StringComparison.Ordinal) ||
+               value.StartsWith("\u03C0", StringComparison.Ordinal) ||
+               value.StartsWith("\u222B", StringComparison.Ordinal) ||
                char.IsDigit(value[0]) ||
-               value.Contains(" – ", StringComparison.Ordinal) ||
+               value.Contains(" - ", StringComparison.Ordinal) ||
                value.Contains("(-", StringComparison.Ordinal) ||
                value.Contains("(+", StringComparison.Ordinal);
     }
 
     private static string CollapseWhitespace(string value) =>
         Regex.Replace(value.Trim(), @"\s+", " ");
+
+    private static string NormalizeText(string value) =>
+        value
+            .Replace('\u00a0', ' ')
+            .Replace('\r', ' ')
+            .Replace('\n', ' ')
+            .Trim();
+
+    private sealed record ColumnAnchor(string Key, double CenterX);
+
+    private sealed record RowAnchor(string Label, int Top, int SortOrder);
 }

src/RolemasterDb.ImportTool/Parsing/XmlTextFragment.cs

@@ -0,0 +1,18 @@
+namespace RolemasterDb.ImportTool.Parsing;
+
+public sealed class XmlTextFragment(
+    int pageNumber,
+    int top,
+    int left,
+    int width,
+    int height,
+    string text)
+{
+    public int PageNumber { get; } = pageNumber;
+    public int Top { get; } = top;
+    public int Left { get; } = left;
+    public int Width { get; } = width;
+    public int Height { get; } = height;
+    public string Text { get; } = text;
+    public double CenterX => Left + (Width / 2.0);
+}