handle accented characters etc

[Rich-Harris]

At present, a file like public/fünke.txt doesn't get served if you hit /fünke, because req.path is /f%C3%BCnke. This fixes it in both dev and prod mode.

[lukeed]

For prod, I think we should just be able encode everything from the glob results, right? Node.js will give us a req.url that's already encoded, so the lookup will pass.

Then for dev, would decode incoming request, which is fine to take the penalty there.

[Rich-Harris]

You've lost me — I thought that's what I'd done in this PR, no?!

[lukeed]

Yeah, definitely! I'm just trying to avoid the decodes on every production request. Node will escape funky characters on their way in.

So I think we can be cheeky and just encode into the cache only, since that's how req.url will already be formed.

[Rich-Harris]

I thought that's what I'm doing by populating FILES with encoded filenames on line 75?

[lukeed]

Well, I feel foolish. Went back to the diff and saw you did exactly what I was describing. I swear the first time around I saw that you were encoding and decoding for prod.

🙈

I've been doing PWA scaffolds for too long. Will merge and release patch in the morning.

Thanks!

[lukeed]

So I'm running this locally with the '/fünke.txt' example you gave. The --dev works – prod has some issues though:

// this is how the "'/fünke.txt'" request comes into Node.js
"/f%C3%BCnke.txt"

// Trying to match that for FILES cache key

encodeURI('/fünke.txt')
'/fu%CC%88nke.txt'

encodeURIComponent('/fünke.txt')
'%2Ffu%CC%88nke.txt'

escape('/fünke.txt')
'/fu%u0308nke.txt'

querystring.escape('fünke.txt')
'fu%CC%88nke.txt'

Likely has to do with the annoying details of unicode that I don't understand... something something multiple codes per character.

Am I missing a possibility? I know decoding works, but I'd like to try paying the toll upfront rather than per request, if possible.

[Rich-Harris]

Huh, that's weird... it was working for me last night. Will try and remember what I did.