Three possible uses for IRIs

[colecmac at protonmail.com]

> (Starting a separate thread for this)

Thanks, that's helpful. Hopefully all IRI discussion can move here.

> Case 3 is the difficult one.  Should authors be allowed to write text/gemini
> links with IRI references? It's not that hard for a client to convert them
> to URI references.  No normalization is needed except as part of punycoding.
> However, everyone has to agree on whether this should work or not; we don't
> want a user trying to follow a link and sending the Wrong Thing to the server.

"It's not that hard for a client to convert them to URI references. No
normalization is needed except as part of punycoding."

I don't think that's true. To convert them to a URI reference, the domain needs
to be extracted and punycoded, then the path and query string needs to be
extracted and percent-encoded in the blessed Gemini way that doesn't allow plus
signs. Doing all this requires parsing, and as I explained a couple times in the
other thread, IRI parsing is not feasible across multiple programming languages
at this time, the libraries just don't exist.

And what if the IRI is a relative reference? As I explained in the other thread,
this will definitely require IRI parsing.

Furthermore, it's breaking change to Gemini. I don't think that's a good idea in
any case with the possible exception of TLS security. Gemini must be reliable,
and it's too late for a breaking change.

> Gemini isn't just supposed to be easy to program for, it's supposed to be easy to
> author, too.  Unfortunately these objectives are in conflict here.

Yes, and that's unfortunate. But I think it makes sense for the stability of Gemini
and the ease of programming to come first.


Cheers,
makeworld