feat(viewer): Revision selection and back-end rewrite

[ketiltrout]

This is a fairly extensive re-write with the following goals:

• make pipeline revision both dynamic and selectable
• remove the front-loading of all a user's opinions on first log-in

Additionally, there's been a bit of tidying up

Removing front-loading

Closes #19

As first written, the viewer used to find all opinions for a user when they first logged in. This data was shoved into the rendered HTML template (in a very non-JSON-y way, for unknown reasons). The only benefit of doing this was that the client never had to contact the server again (except if the user wanted to update their opinion).

Fine in theory, if there are only a few opinions for a user (which was true when the viewer was first written), but causes more-and-more slowdown as opinions accumulate over time.

In place of this, the client now side-loads opinions for a CSD (for all available revisions) by sending a fetch request to the server, which returns the data in JSON (see fetch_csd in the server, though most of the work is done in csd_data; and also set_csd in the client).

The downside here is there is a delay when switching CSDs in the viewer as the viewer has to wait for the server to return the data, but I don't find this delay noticable in practice.

Specific server-side changes:

• NB: the "fetch" query parameter was previously used when the client asked for a rendered notebook page. That key has been renamed to "render". The "fetch" parameter is now used for a CSD data side-load request
• The fetch_csd function is used to respond to these side-loading requests. Most of the actual data generation is implemented in csd_data (which evolved from the csd_vars function and is also used after an opinion update to generate the data returned to the client).

Specific client-side changes:

• The csd_list global has been deleted (formerly a list of all available CSDs)
• The opinions global now holds the user's opinion for the current CSD only (indexed by revision), rather than all CSDs.
• The vet_csd function, which was used to validate numbers typed into the CSD box, has been deleted (the functionality is now handled by the server during the side-loading)

Pipeline revision selection

Closes #20

The fixed global _REVISION has been removed. In its place, the viewer now will dynamically figure out which revisions to load/display:

• On first load (i.e. when a worker first serves a page and loads its globals, the server walks through DataRevision and finds all revisions for which at least one rendered notebook exists. This ends up as the REVISIONS global, and will remain fixed until server restart.
• Due to the side-loading, whenever data from a new CSD is requested, the server will find all rendered notebooks for that CSD and provide the data to the client.
• In the client there are now buttons to switch between revisions. The list of buttons is fixed by the contents of REVISIONS (i.e. it doesn't change unless a server restart happens) but, if the currently-viewed CSD has no rendered notebook for a particular revision, the corresponding revision button is disabled to indicate that.
• Otherwise, revision buttons are coloured to indicate the user's submitted opinion for that day (if any).
• Because a CSD-fetch side-load returns all revisions, switching between revisions for a CSD does not cause another server hit.

Specific changes:

• revision number (rev parameter) must be submitted with all requests now, and is provided in all responses. When not given a valid revision number, the server will default to the lowest available revision. Instances where a client's query data was stripped of "unimportant" data has now been unified into the strip_query function.
• The update_ui client function has been broken up into two parts: update_data which updates the globals based on a server's response, and draw_ui which performs the document update based on the state of the globals. update_data is called by side-loader callbacks, and in turn will call draw_ui, but the draw_ui function is also called by the revision selection buttons (via their set_rev function), which doesn't involve a side-load.

Other changes

I've removed the script_name and base_path vars which were an unnecessary attempt to tell the client where to send requests: the client doesn't need this; it already knows the URL it's using. More evidence that these are not needed: they were broken when we transitioned the viewer to gunicorn (both ended up as "") but the client didn't notice or care).

This PR breaks up the update_opinion function in the server, which isn't technically necessary, and only done in anticipation of the category handling stuff.

[ketiltrout]

I've left the test-version running here (with only a few CSDs available) if you want to test it: https://bao.chimenet.ca/daily-test/view

[ketiltrout]

I've realised that the CSD selection buttons don't update as they should when revision change happens in the UI. I don't think this is a big enough problem to fix in this PR (since it only affects the CSD you're currently on when you revision switch - as soon as you navigate to another CSD things are fixed), but I've made an issue for it (#26) and I'll deal with it in a subsequent fix.

[ketiltrout]

I think the code all seems fine. The test viewer seems to work without issue, so I think this is stable enough to deploy.

This is incredibly minor, but in safari the buttons change shape depending on whether or not there's an opinion-based colour being set.

Hmm. I'm changing the border property. Maybe I can just change the border colour without affecting the rest of the border styling.

[ljgray]

Hmm. I'm changing the border property. Maybe I can just change the border colour without affecting the rest of the border styling.

If it doesn't end up being something simple to debug, I don't think it should block us from deploying this update

[ketiltrout]

In the end, I just disabled webkit styling from those buttons. So they look consistent, at least.