How #Seppo! works

#Seppo! is meant for people[1] to operate their own, first-person (single-user) microblog. Therefore gentle onboarding and safe operation are paramount. And because server operation has it's pitfalls, #Seppo! encourages to rent cheap shared webspace and this way have professionals do the delicate but standard tasks (🏡 Webspace) on an affordable but sustainable business perspective.

You do not need privileged server access ('root') to have your own #Seppo! but you do need a CGI[2]-friendly webserver. E.g. nginx or caddy aren't.

Setup

#Seppo! is a CGI program driven by a webserver, e.g. Apache from a LAMP[3]-stack. Put seppo.cgi (see 🌱 Installation) on your CGI-capable shared webspace and enter the address in your browser's address bar, e.g. https://example.org/sub/dir/seppo.cgi. Being brand-new, the CGI doesn't find the marker file delete-me-to-restore-missing, unpacks all assets (xsl[4], css, delete-me-to-restore-missing, etc.) and creates private and public facing files, finds no app/etc/passwd.s either and so asks to set login credentials. Then it tries to create a symlink in the webspace's root /.well-known/webfinger/.htaccess. Only if self-hosting or not on Apache, you have to care about the webserver configuration. There are config file samples in contrib/etc/[5].

Discovery

In order to be found via WebFinger[6], a http GET request to https://example.org/.well-known/webfinger?resource=acct:alice@example.org must respond with the proper JRD[7] document.

#Seppo! sets this up automatically, see S1002 for details.

Add a Post

For now via web-interface. Posting does:

1. append the message to the global app/var/db/o/p.s
2. append to the most recent index page app/var/db/o/p/0.s or start a new one
3. re-create the public facing according blog page o/p-0/index.xml
4. evtl. change the symlink to the 'current' blog page o/p/index.xml
5. do likewise for tags o/t/*-0/index.xml and daily o/d/*-0/index.xml pages
6. queue a job to notify each follower in app/var/spool/job/new/
7. trigger queue processing and loop via minutely HTTP calls to itself

Delete a Post

Typically via web-interface. Deleting does:

1. find the entry in the primary page index app/var/db/o/p/0.s
2. overwrite the post in the global app/var/db/o/p.s
3. re-create the public facing according blog page o/p-0/index.xml
4. re-create the according tags and daily overview pages
5. queue a job to notify each follower in app/var/spool/job/new/
6. trigger queue processing and loop via minutely HTTP calls to itself

Incoming followed Posts

Incoming activities[8]:

1. verify signature[9],
2. match sending actor[10],
3. parse[11],
4. serialize and purge after 90 days.

Design Decisions

To work unattended, #Seppo! avoids risks wherever possible. Namely

• do as little as possible (Occams Razor)
• prefer safe over convenient or quick
• have a 2sec sleep on all login attempts
• ban after repeated failed logins
• never completely fill the disk
• do log swapping, not rotation
• have a timeout/max retries on everything
• all persistence to inspectable (mostly text) files
• Canonical S-Expressions for storage, no json nor xml
• no mandatory scripting, neither server nor client
• minimal Trusted Computing Base, few dependencies, both toolchain and runtime, no database, no scripting runtime, few small & seasoned libraries (e.g. tls[12], jsonm[13], xmlm[14], lambdasoup[15]) lean on transitive dependencies and from trusted authors (Mehnert, Bünzli, Bachin)
• only accept parsed and valid external data and never redistribute unattended
• accept incoming posts only from people followed
• best practices (e.g. don't parse, make undesired state unrepresentable, immutability changes everything)
• a memory-safe programming language, 🐫 OCaml[16] with a competent, friendly community
• render html client-side via xsl
• built on solid web foundations, not the latest fads (Linux, Apache, CGI[2], Atom[17], XSL[4], XHTML[18], CSS)

* * *

1. The Internet is for End Users (RFC8890), back^
2. The Common Gateway Interface (CGI) (RFC3875), back^, back^
3. LAMP (software_bundle), #Seppo needs the L and A, back^
4. XSL Transformations (XSLT), back^, back^
5. contrib/etc/, back^
6. WebFinger (RFC7033), back^
7. 4.4 JSON Resource Descriptor, back^
8. 3.1 Activity Types, back^
9. 2.5. Verifying a Signature, back^
10. 3.2 Actor Types, back^
11. https://codeberg.org/seppo/seppo/src/commit/973d3bb/as2_vocab/decode.ml#L380-L384, back^
12. tls, back^
13. jsonm, back^
14. xmlm, back^
15. lambdasoup, back^
16. 🐫 OCaml, back^
17. Atom (RFC4287, RFC4685, RFC5005), back^
18. XHTML™ 1.1 - Module-based XHTML - Second Edition, back^