Should Snowpack stop reading binary files?

[drwpow]

In snowpack build, we pass binary files to load() and transform() in plugins, as well as try and parse those files for imports to install. But should we stop doing this, and just decide that if a file is binary, we don’t send it to any plugins?

The pros to this approach would be performance and memory usage: by simply skipping many files, we could deal with less headache and memory problems by not loading potentially-gigantic files in memory.

But the cons would be utility—by omitting binary files it would mean that if people wanted to do some kind of transformation to binary files (e.g. image optimizer), that would have to all happen at the bundle() step, which means you couldn’t manipulate binary files for the dev server at all.

It should be noted that this is a companion discussion to the work happening here: #980

Any thoughts?

cc @gr2m @FredKSchott

[gr2m]

what would be a use case to manipulate binary files for the dev server?

[drwpow]

Maybe there’s a Snowpack plugin that generates LQIP (low-quality image placeholders) or adds WebP support or something. May be helpful to test that in dev.

Beyond that, I can’t really think of anything.

[gr2m]

If the memory usage is a concern, we could make the .contents property a getter, which would only read out the file content when needed

[FredKSchott]

A few thoughts:

• In dev, the nice thing is that "if it's being sent through the build pipeline, it's being loaded in the browser, so it can't be THAT huge". But that doesn't help us during build...
• load() is actually okay: it only gets the file name AND it would only be triggered if you specifically used a binary file extension in your resolve config. So this is really a transform() issue.
• breaking change fears make me feel like since this is already behavior, we shouldn't mess with it. I believe that there's already a plugin that works with issues (can't remember it tho, I could be mistaken).