Ank.DocToolkit.Extensions.DependencyInjection 0.10.0

Ank.DocToolkit.Extensions.DependencyInjection

Dependency-injection registration for Ank.DocToolkit — services.AddDocToolkit() registers six injectable interfaces over the same pure-managed HTML/DOCX/PDF/XLSX/PPTX conversion and editing logic.

dotnet add package Ank.DocToolkit.Extensions.DependencyInjection

Targets net8.0 and net10.0. MIT licensed.

Usage

using DocToolkit.Extensions.DependencyInjection;

services.AddDocToolkit();

// Or opt in to remote image download for HTML->DOCX/PDF. This still succeeds in an air-gapped
// environment - an unreachable host leaves that image out rather than failing the conversion.
services.AddDocToolkit(o => o.AllowRemoteImageDownload = true);

Bounding the remote-image opt-in

AllowRemoteImageDownload is the only switch that decides whether anything is fetched. When it is true, every fetch is bounded by RemoteImage, whose defaults are already the restrictive ones — loopback, private and link-local addresses are refused (including 169.254.169.254, the cloud metadata endpoint), only http and https are spoken, redirects are not followed, and each fetch is capped at 10 seconds and 5 MB counted on bytes actually read.

services.AddDocToolkit(o =>
{
    o.AllowRemoteImageDownload = true;
    o.RemoteImage.Timeout = TimeSpan.FromSeconds(3);
    o.RemoteImage.AllowedHosts.Add("cdn.example.com");   // empty means "any public host"
});

RemoteImage is configured in place, not assigned: the property is get-only so that a restrictive default cannot be lost by dropping in an object that missed one.

Fetching from an intranet image host? The address block refuses private ranges by default, so that image is skipped silently. Set o.RemoteImage.AllowPrivateAddresses = true to allow it — and be aware that doing so is what re-opens the SSRF reach if any caller converts untrusted HTML.

This is not a complete SSRF defence. A host's address is resolved and checked, then resolved again by the HTTP stack when it connects; a DNS answer that changes in between defeats the check. See the core package README and SECURITY.md.

public class InvoiceService
{
    private readonly IHtmlToDocxConverter _toDocx;
    private readonly IHtmlToPdfConverter _toPdf;

    public InvoiceService(IHtmlToDocxConverter toDocx, IHtmlToPdfConverter toPdf)
    {
        _toDocx = toDocx;
        _toPdf = toPdf;
    }

    public Task<byte[]> RenderAsync(string html) => _toPdf.ConvertAsync(html);
}

// Every interface also has Stream-based async members, so a large document never has to be
// duplicated into a caller-visible byte[] — write straight to an HTTP response body instead:
record InvoiceRequest(string Html);

app.MapPost("/invoices/pdf", async (InvoiceRequest request, IHtmlToPdfConverter toPdf, HttpResponse response) =>
{
    response.ContentType = "application/pdf";
    // The PDF is written to response.Body as it is rendered, not assembled first, so the
    // status code and headers are committed on the first write. A failure part-way through
    // cannot be turned into a clean 500 — the response is already underway.
    await toPdf.ConvertAsync(request.Html, response.Body);
});

All six interfaces — IHtmlToDocxConverter, IDocxToPdfConverter, IHtmlToPdfConverter, IDocxEditor, IWorkbookEditor, IPresentationEditor — mirror Ank.DocToolkit's static API, including both its byte[] and its Stream-based async overloads. They are registered as singletons (each wraps stateless logic) and are safe to inject and call concurrently. See the core package's README for what each one does and the offline/licensing guarantees behind them.

Two things on the static API deliberately do not appear on these interfaces:

• The file-path helpers (ConvertToFileAsync, ConvertFile). Inject the converter, take the byte[] or write to a Stream, and put the bytes wherever they belong — that keeps the injected surface free of filesystem coupling.
• The per-call allowRemoteImageDownload argument and RemoteImageOptions overloads. Remote image download is configured once, at registration, via DocToolkitOptions — so whether an application may reach the network, and how far, is a property of how it is composed rather than a decision at each call site. It is false unless you opt in.

Why a separate package

A console app, Lambda or simple script that only wants the static byte[]-based API installs just Ank.DocToolkit, with zero DI dependencies. ASP.NET Core and worker-service consumers add this package too.

Licence

MIT — see the parent repository's LICENSE.