Add crawling TODOs

[tt.git] / TODO

# vim:sw=2:sts=2:
TODO
====

Legend:
- [ ] not started
- [-] in-progress
- [x] done
- [~] cancelled

In-progress
-----------
- [-] Convert to Typed Racket
  - [x] build executable (otherwise too-slow)
  - [-] add signatures
    - [x] top-level
    - [ ] inner
    - [ ] imports
- [-] commands:
  - [x] c | crawl
    Discover new peers mentioned by known peers.
  - [x] r | read
    - see timeline ops above
  - [ ] w | write
    - arg or stdin
    - nick expand to URI
    - Watch FIFO for lines, then read, timestamp and append [+ upload].
      Can be part of a "live" mode, along with background polling and
      incremental printing. Sort of an ii-like IRC experience.
  - [ ] q | query
    - see timeline ops above
    - see hashtag and channels above
  - [x] d | download
    - [ ] options:
      - [ ] all - use all known peers
      - [ ] fast - all except peers known to be slow or unavailable
        REQUIRES: stats
  - [x] u | upload
    - calls user-configured command to upload user's own timeline file to their server
  Looks like a better CLI parser than "racket/cmdline": https://docs.racket-lang.org/natural-cli/
  But it is no longer necessary now that I've figured out how to chain (command-line ..) calls.
- [-] Output formats:
  - [x] text long
  - [x] text short
  - [ ] HTML
  - [ ] JSON
- [-] Peer discovery
  - [-] parse peer refs from peer timelines
    - [x] mentions from timeline messages
      - [x] @<source.nick source.url>
      - [x] @<source.url>
    - [ ] "following" from timeline comments: # following = <nick> <uri>
  - [ ] Parse User-Agent web access logs.
  - [-] Update peer ref file(s)
    - [x] peers-all
    - [x] peers-mentioned
    - [ ] peers-followed (by others, parsed from comments)
    - [ ] peers-down (net errors)
    - [ ] redirects?
    Rough sketch from late 2019:
        let read file =
            ...
        let write file peers =
            ...
        let fetch peer =
            (* Fetch could mean either or both of:
             * - fetch peer's we-are-twtxt.txt
             * - fetch peer's twtxt.txt and extract mentioned peer URIs
             * *)
            ...
        let test peers =
            ...
        let rec discover peers_old =
            let peers_all =
                Set.fold peers_old ~init:peers_old ~f:(fun peers p ->
                    match fetch p with
                    | Error _ ->
                        (* TODO: Should p be moved to down set here? *)
                        log_warning ...;
                        peers
                    | Ok peers_fetched ->
                        Set.union peers peers_fetched
                )
            in
            if Set.empty (Set.diff peers_old peers_all) then
                peers_all
            else
                discover peers_all
        let rec loop interval peers_old =
            let peers_all = discover peers_old in
            let (peers_up, peers_down) = test peers_all in
            write "peers-all.txt"  peers_all;
            write "peers-up.txt"   peers_up;
            write "peers-down.txt" peers_down;
            sleep interval;
            loop interval peers_all
        let () =
            loop (Sys.argv.(1)) (read "peers-all.txt")

Backlog
-------
- [ ] Crawl all cache/objects/*, not given peers.
  BUT, in order to build A-mentioned-B graph, we need to know the nick
  associated with the URI whos object we're examining. How to do that?
- [ ] Crawl downloaded web access logs
- [ ] download-command hook to grab the access logs

    (define (parse log-line)
      (match (regexp-match #px"([^/]+)/([^ ]+) +\\(\\+([a-z]+://[^;]+); *@([^\\)]+)\\)" log-line)
        [(list _ client version uri nick) (cons nick uri)]
        [_ #f]))

    (list->set (filter-map parse (file->lines "logs/combined-access.log")))

    (filter (λ (p) (equal? 'file (file-or-directory-type p))) (directory-list logs-dir))

- [ ] user-agent file as CLI option - need to run at least the crawler as another user
- [ ] Support fetching rsync URIs
- [ ] Check for peer duplicates:
  - [ ] same nick for N>1 URIs
  - [ ] same URI for N>1 nicks
- [ ] Background polling and incremental timeline updates.
  We can mark which messages have already been printed and print new ones as
  they come in.
  REQUIRES: polling
- [ ] Polling mode/command, where tt periodically polls peer timelines
- [ ] nick tiebreaker(s)
  - [ ] some sort of a hash of URI?
  - [ ] angry-purple-tiger kind if thingie?
  - [ ] P2P nick registration?
    - [ ] Peers vote by claiming to have seen a nick->uri mapping?
          The inherent race condition would be a feature, since all user name
          registrations are races.
          REQUIRES: blockchain
- [ ] stats
  - [ ] download times per peer
- [ ] Support redirects
  - should permanent redirects update the peer ref somehow?
- [ ] Support time ranges (i.e. reading the timeline between given time points)
- [ ] optional text wrap
- [ ] write
- [ ] timeline limits
- [ ] peer refs set operations (perhaps better done externally?)
- [ ] timeline as a result of a query (peer ref set op + filter expressions)
- [ ] config files
- [ ] highlight mentions
- [ ] filter on mentions
- [ ] highlight hashtags
- [ ] filter on hashtags
- [ ] hashtags as channels? initial hashtag special?
- [ ] query language
- [ ] console logger colors by level ('error)
- [ ] file logger ('debug)
- [ ] Suport immutable timelines
  - store individual messages
    - where?
      - something like DBM or SQLite - faster
      - filesystem - transparent, easily published - probably best
  - [ ] block(chain/tree) of twtxts
    - distributed twtxt.db
    - each twtxt.txt is a ledger
    - peers can verify states of ledgers
    - peers can publish known nick->url mappings
    - peers can vote on nick->url mappings
    - we could break time periods into blocks
    - how to handle the facts that many(most?) twtxt are unseen by peers
    - longest X wins?

Done
----
- [x] Dedup read-in peers before using them.
- [x] Prevent redundant downloads
  - [x] Check ETag
  - [x] Check Last-Modified if no ETag was provided
    - [x] Parse rfc2822 timestamps
- [x] caching (use cache by default, unless explicitly asked for update)
  - [x] value --> cache
  - [x] value <-- cache
    REQUIRES: d command
- [x] Logger sync before exit.
- [x] Implement rfc3339->epoch
- [x] Remove dependency on rfc3339-old
- [x] remove dependency on http-client
- [x] Build executable
  Implies fix of "collection not found" when executing the built executable
    outside the source directory:

    collection-path: collection not found
      collection: "tt"
      in collection directories:
      context...:
       /usr/share/racket/collects/racket/private/collect.rkt:11:53: fail
       /usr/share/racket/collects/setup/getinfo.rkt:17:0: get-info
       /usr/share/racket/collects/racket/contract/private/arrow-val-first.rkt:555:3
       /usr/share/racket/collects/racket/cmdline.rkt:191:51
       '|#%mzc:p


Cancelled
---------
- [~] named timelines/peer-sets
  REASON: That is basically files of peers, which we already support.