TLS close-notify .... what were they thinking?

=> http://garrett.damore.org/2017/11/tls-close-notify-what-were-they-thinking.html

At least the new Gemini spec considers the case where servers choose to skip this step. How do you handle a missing close_notify in your clients?

=> Posted in: s/Gemini | 🛰️ lufte

2024-12-13 · 7 weeks ago

5 Comments ↓

=> 🚀 stack · Dec 13 at 18:52:

what where they thinking

My pet theory explaining this as well as the insane gemtext line type markers is drug-induced psychosis. It gets some people angry, but it's based on personal experience.

As they say, "you had one job..."

=> 👻 ps · Dec 13 at 18:59:

In Rust, using Gio SocketConnection - it closing anyway, on reference count get 0 lol.

Personally, don't see any sense to keep it active as can't send new request to existing output stream. For media streams, it could be another story, of course.

=> 🦂 zzo38 · Dec 14 at 03:47:

The "astroget" program does not currently handle close notify (in future it might be, although I had not tried to figure out how).

Presumably, TLS close notify is helpful for Gemini to indicate EOF, since there is no length header. (Even if there is a length header, the server might not always know the length of the file, e.g. in case it is a dynamic file, so it cannot always be used.)

=> 😈 dimkr · Dec 15 at 07:31:

My client has this horrible line (https://github.com/dimkr/gplaces/blob/f7254ed468effb9912a56db92052ef04791527ec/gplaces.c#L943) and I'd love to remove it but last time I checked there were still popular capsules (one of the search engines I think?) that don't send close_notify

=> 🧇 Acidus · Dec 20 at 03:59:

Until a month or so ago my search engine Kennedy didn't cleanly close the TLS connection and send closed_notify. @omar pointed it out to me and I fixed!