This example builds both tables in C#. It does not open DOCX or spreadsheets.
For digital PDFs, use RecursiveAI.OneDiff.Pdf 1.0.1. For DOCX, build a CompareDocument with your own reader; no released OneDiff package reads DOCX.
dotnet add package RecursiveAI.OneDiff.DocumentsRequires .NET 10 (net10.0); .NET 9 and earlier aren’t supported.
using OneDiff.Documents; using OneDiff.Documents.Model; var result = Comparison.Of( Table("Pending"), Table("Approved")); foreach (var edit in result.Changes) { var path = edit.New.Anchor.Reference; Console.WriteLine(path); Console.WriteLine( $"{edit.Old.Text} → " + edit.New.Text); } static CompareDocument Table(string status) { var doc = new CompareDocumentBuilder(); doc.Begin(NodeKind.Table); doc.Begin(NodeKind.TableRow); doc.Begin(NodeKind.TableCell); doc.Add(NodeKind.Paragraph, new() { Text = status, Anchor = Anchor.ForNode( "orders/42/status") }); doc.End(); // cell doc.End(); // row doc.End(); // table return doc.Build(); }
dotnet run prints:
orders/42/status Pending → Approved
Begin opens a container; End closes it. An anchor is a label you choose for each cell’s location, such as orders/42/status. Put it on the paragraph that holds the cell’s text. For an inserted cell, Old.IsPresent is false and Old.Text is empty; for a deleted cell, New.IsPresent is false and New.Text is empty. To read the path from the side that exists, replace the path assignment in the loop with:
var location = edit.New.IsPresent ? edit.New : edit.Old; var path = location.Anchor.Reference;
An inserted or deleted cell can produce separate changes for the cell container and its text. The container has no path in this example; the text change carries the anchor.
Unzip the example, then dotnet run.
Variations
To see where a change came from as well as where it landed, replace Console.WriteLine(path); in the loop with the following lines, then remove the now-unused path variable:
Console.WriteLine(edit.Old.Anchor.Reference); Console.WriteLine(edit.New.Anchor.Reference);
Both lines print orders/42/status here because the cell kept its path.
Related recipes
The short answer
Build each version as a CompareDocument with table, row, and cell nodes. Put the cell text in a child paragraph and give that paragraph a source path using Anchor.ForNode. After Comparison.Of, read the change’s old and new anchors and text. Here, orders/42/status identifies order 42’s status cell, which changed from Pending to Approved.
How do you compare two tables and identify differences?
Keep the table hierarchy rather than comparing a flat list of lines. A table groups rows, a row groups cells, and a cell can group content such as paragraphs. The container nodes in the example have no text of their own. Only the paragraph contains the word that changes.
The alignment rules use matched content to pair containers. They can also pair containers in order when equal numbers of unmatched containers of the same kind fall between matching siblings. Within a paired cell, a single unmatched text block of the same kind on each side can then be paired even when its wording is completely different. That’s why Pending and Approved show up together rather than as one deletion and one insertion in the sample.
Which path belongs in a change report?
Use each side’s own path. A node number is its position in that version’s document tree, so inserting a row can change numbers downstream. A path you generate from a persistent row identifier and a column name can stay useful even when rows move. If the source only has row and column positions, record the positions separately for each version rather than treating them as permanent identities.
A source path tells the reader where a change occurred; it does not tell OneDiff which rows to match. The compare model describes an anchor as “a source location the engine copies but never interprets.” If you need a strict join by order number, do that pairing in your application instead of expecting the path string to act as a join key.
This is a comparison of document content you supply, not a workbook comparison. The 1.0 packages do not read grids, and the sample does not infer formulas or cell addresses from display text.