What is this Gemini thing anyway, and why am I excited about it? November 1, 2020 on Drew DeVault's blog

I’ve been writing about some specific topics in the realm of Gemini on my blog over the past two months or so, but I still haven’t written a broader introduction to Gemini, what I’m doing with it, and why you should be excited about it, too. Let’s do that today!

Gemini is a network protocol for exchanging hypertext documents — “hypertext” in the general sense of the word, not with respect to the hypertext markup language (HTML) that web browsers understand. It’s a simple network protocol which allows clients to request hypertext documents (in its own document format, gemtext). It is, in some respects, an evolution of Gopher, but more modernized and streamlined.

Gemini is very simple. The protocol uses TLS to establish an encrypted connection (using self-signed certificates and TOFU rather than certificate authorities), and performs a very simple exchange: the client sends the URL it wants to retrieve, terminated with CRLF. The server responds with an informative line, consisting of a numeric status code and some additional information (such as the document’s mimetype), then writes the document and closes the connection. Authentication, if desired, is done with client certificates. User input, if desired, is done with a response code which conveys a prompt string and a request for user input, followed by a second request with the user’s response filled into the URL’s query string. And that’s pretty much it!

$ openssl s_client -quiet -crlf   \
    -servername drewdevault.com   \
    -connect drewdevault.com:1965 \
  | awk '{ print "response: " $0 }'
gemini://drewdevault.com
response: 20 text/gemini
response: ```ASCII art of a rocket next to "Drew DeVault" in a stylized font
response:   /\
response:   ||    ________                         ________       ____   ____            .__   __
response:   ||    \______ \_______   ______  _  __ \______ \   ___\   \ /   /____   __ __|  |_/  |_
response:  /||\    |    |  \_  __ \_/ __ \ \/ \/ /  |    |  \_/ __ \   Y   /\__  \ |  |  \  |\   __\
response: /:||:\   |    `   \  | \/\  ___/\     /   |    `   \  ___/\     /  / __ \|  |  /  |_|  |
response: |:||:|  /_______  /__|    \___  >\/\_/   /_______  /\___  >\___/  (____  /____/|____/__|
response: |/||\|        \/            \/                 \/     \/             \/
response:   **
response:   **
response: ```
[...]

So why am I excited about it?

My disdain for web browsers is well documented¹. Web browsers are extraordinarily complex, and any attempt to build a new one would be a Sisyphean task. Successfully completing that implementation, if even possible, would necessarily produce a Lovecraftian mess: unmaintainable, full of security vulnerabilities, with gigabytes in RAM use and hours in compile times. And given that all of the contemporary web browsers that implement a sufficiently useful subset of web standards are ass and getting assier, what should we do?

The problem is unsolvable. We cannot have the “web” without all of these problems. But what we can have is something different, like Gemini. Gemini does not solve all of the web’s problems, but it addresses a subset of its use-cases better than the web does, and that excites me. I want to discard the parts of the web that Gemini does better, and explore other solutions for anything that’s left of the web which is worth keeping (hint: much of it is not).

There are some aspects of Gemini which I approve of immensely:

• It’s dead simple. A client or server implementation can be written from scratch by a single person in the space of an afternoon or two. A new web browser could take hundreds of engineers millions of hours to complete.
• It’s not extensible. Gemini is designed to be difficult to extend without breaking backwards compatibility, and almost all proposals for expansion on the mailing list are ultimately shot down. This is a good thing: extensibility is generally a bad idea. Extensions ultimately lead to more complexity and Gemini might suffer the same fate as the web if not for its disdain for extensions.
• It’s opinionated about document formatting. There are no inline links (every link goes on its own line), no formatting, and no inline images. Gemini strictly separates the responsibility of content and presentation. Providing the content is the exclusive role of the server, and providing the presentation is the exclusive role of the client. There are no stylesheets and authors have very little say in how their content is presented. It’s still possible for authors to express themselves within these constraints — as with any other constraints — but it allows clients to be simpler and act more as user agents than vendor agents.

Some people argue that what we should have is “the web, but less of it”, i.e. a “sane” subset of web standards. I don’t agree (for one, I don’t think there is a “sane” subset of those standards), but I’ll save that for another blog post. Gemini is a new medium, and it’s different from the web. Anyone who checking it out should be prepared for that and open to working within its constraints. Limitations breed creativity!

For my part, I have been working on a number of Gemini projects. For one, this blog is now available on Gemini, and I have started writing some Gemini-exclusive content for it. I’ve also written some software you’re welcome to use:

libgmni, gmni, and gmnlm are my suite of Gemini client software, all written in C11 and only depending on a POSIX-like system and OpenSSL. libgmni is a general-purpose Gemini client library with a simple interface. gmni is a cURL-like command line tool for performing Gemini requests. Finally, gmnlm is a line-mode browser with a rich feature-set. Together these tools weigh just under 4,000 lines of code, of which about 1,600 are the URL parser from cURL vendored in.

gmnisrv is a high-performance Gemini server, also written in C11 for POSIX systems with OpenSSL. It supports zero-configuration TLS, CGI scripting, auto-indexing, regex routing and URL rewrites, and I have a couple more things planned for 1.0. It clocks in at about 6,700 lines, of which the same 1,600 are vendored from cURL, and an additional 2,800 lines are vendored from Fabrice Bellard’s quickjs regex implementation.

kineto is an HTTP-to-Gemini gateway, implemented as a single Go file (under 500 lines) with the assistance of ~adnano’s go-gemini library. My Gemini blog is available through this portal if you would like to browse it.

So dive in and explore! Install gmnisrv on your server and set up a Gemini space for yourself. Read the feeds from CAPCOM. Write some software of your own!