Offline jEdit trackers browsing using kinto and kinto.js
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
config
.gitignore
README.md
bower.json
imap_socketio.py
requirements.txt
socketio_updater.py
testsocketio.html
trackers.css
trackers.html
trackers.js
transform.js
updater.py
updater.sh

README.md

Browsing jEdit Trackers with kinto.js: An experiment

1. What I wanted to achieve

Access the jEdit sourceforge bug trackers while I'm off the internet.

It's been solved before, using dumps from Sourceforge: http://jedit.org/trackers/ provides a downloadable archive of html+a little js to make browsing more convenient (it's also browsable online).

The problem for me is I have to remember to run the update script (has my sourceforge credentials) every week or so (the dumps are taking minutes to generate, so I don't run it so often).

The problem for users is they have to re-download the whole archive each time (i.e. no auto-update).

Try out Kinto

It seems interesting:

  • kinto has a clean API;
  • it has learned from couchDB;
  • kinto.js taps into the browsers' indexedDB and lets you synchronize with the server's database;
  • each user may create his collections/records, opening an application for extensions, as well as define fine grained permissions.

2. What I have

Server

A kinto server is running at https://kinto.elelay.fr. I'm still in the process of setting it up. It's running in docker behind an nginx proxy.

I've followed the permission tutorial to make 2 read-only collections in the jedit-trackers bucket (tickets and oldTickets).

Client

A working application is at https://trackers.elelay.fr/trackers.html

It will let you download and browse tickets in a table.

Search is done by a combination of quick filters and full text search:

  • one of
    • open tickets in different trackers
    • missed (open tickets without answer older than 2 weeks)
    • all (include old tickets in the query)
  • combined with full-text-search in summary and description

A result can be expanded, or opened (in a new tab using middle-mouse-button click). One can jump to Sourceforge to edit the ticket.

Search queries are bookmarkable, thanks to an ugly encoding of the state object in the hash. Tickets are also bookmarkable and a bit prettier (id in the hash).

It polls the Kinto server for updates every minute, and will prompt you to update when tickets are updated.

Counters and the full-text-search index are saved in a local stats collection.

It uses

  • jQuery Datatables
  • lodash, moment
  • bootstrap's CSS, fontawesome icons
  • the lunr package for client-side full-text search

It doesn't use any particular client framework, and the code grew organically to just under 1k lines.

Updates

  • I subscribed to jEdit-devel mailing list via a dedicated address and monitor ticket-related mail notifications in imap_socketio.py;
  • it publishes ticket events via a socket.io endpoint.
  • socketio_updater.py then subscribes to this websocket and fetches and update tickets in the Kinto database. I get almost real-time updates when sending test emails to the monitoring address. We'll see how it fares with real SF notifications.

I run updater.py on the kinto server, polling every day for changes in Sourceforge trackers in case of missed updates.

3. Improvements

I want to let users write to tickets: ATM they simply fetch a read-only DB, and kinto can do more. Maybe not write to the tickets per se, but maintain a user collection with stars, tags, etc..

I initially wanted to apply changes in Sourceforge from forms in the application, using the REST API and XMLHTTPRequests. But it's a lot of work to create forms and the like.

I envisioned these changes would be stored in a separate collection and applied when network would be available. The user would have to login using Sourceforge's oauth flow. Apparently it's not a good idea to do it client side because the application secret token is leaked to everybody. Maybe use the bearer token for now.

The interface is not ugly (I think), but feels clunky.

Picking a client-side framework (angular-js, react) would help me organise the code but I wanted to be close to vanilla.

Attachments: who doesn't like attachments? I've not had time to look into kinto's attachments, but it's on my todo. Until that, the offline experience may be a bit unperfect: you find an interesting bug to work on, browse the discussion, stumble upon the inevitable "please upload the activity.log", see the OP has provided it, but the link doesn't work, because you're off the internet :-( So I'll try to fetch all attachments and put them in the kinto server. I have no idea of their size yet...

I'm not using lunr.js to it's full potential by indexing so little fields. The index is already a very demanding JSON blob to load and save. I'm in two minds about that: expand the index by feeding it everything, enabling richer queries or find another way to do full-text search.

Indexes are not available yet. They could help make some queries quicker (like missed tickets). I'm considering saving indexes myself in the stats collection (for instance a document with all ticket ids by num).

Full-text indexing is a good fit for a webworker. I'll have to look into that...