Image processing as metadata

[olevett]

I have a pipeline setup that I've migrated over from Wyam to Statiq that has worked fine, until the latest beta versions which I think change some of the IO processing. I've been feeling like it's a little hacky anyway, and thought I'd ask a question of it and see where it goes...

I have a few different document types for different pages which share a common "header-image" metadata field which points to a file path, e.g.

header-image: assets/header-01.png
---
Some content

which can be edited through Netlify. Because the images are uploaded by content editors, they can sometimes end up large/the wrong aspect ratios, so I've been resizing them, and then updating the document metadata with the new file path, with something like this:

public static async Task<IDocument> CopyAndResizeImageFromField(
            IDocument doc,
            IExecutionContext ctx,
            string fieldName,
            int? width,
            int? height,
            ResizeMode resizeMode = ResizeMode.Min)
{
    var fileName = doc.Get<string>(fieldName).TrimIfStartsWith("/");
    var documents = await ctx.ExecuteModulesAsync(
        new ModuleList
        {
            new ReadFiles(fileName),
            new MutateImage().Resize(width: width, height: height, mode: resizeMode),
            new WriteFiles()
        },
        new IDocument[] { doc }
        );
    return documents.FirstOrDefault();
}

// ...
// in a pipeline
ProcessModules = new ModuleList
{
// ...
   new SetMetadata("header-image", Config.FromDocument((doc, ctx) => HeaderImageExtensions.CopyAndResizeImageFromField(doc, ctx, "header-image", 1280, null, ResizeMode.Pad))),
// ...
};

which

• grabs the image path
• runs it through an image resize process
• returns a new document with the new file path in the output
  and then this gets written out in Razor later as an <img>

All of this worked fine, but felt a little icky (e.g. the call to ExecuteModulesAsync and the having an inner module do a file output), but this on the latest betas is now failing if multiple input documents point to the same header images, and try to write out to the same file, there's a conflict at the file system level.

There are a couple of things I've considered, but wonder if there's a step back that would make it more sensible, or if there's something I've missed?

• memoize the CopyAndResizeImageFromField so there's only ever one output for the same input document
• output a random file name so there's no conflict at the output stage
• pull out a separate pipeline that just does header images for all input documents, the sticking point here is how to mutate the documents to have the updated file names from the post processing