rethink partial tiles

[jfirebaugh]

The logic around partial tiles is complex and brittle. We need to rethink this, maybe do it in a totally different way.

cc @tmpsantos @kkaefer

[kkaefer]

What aspect of it is causing problems for you?

[jfirebaugh]

Our current implementation is highly stateful, and it is difficult to reason about the order or timing of state transitions. This has given rise to pattern of issues where tiles either never finish parsing or are parsed ad infinitum:

• rendering tests: Error: timed out after 20 seconds #2808
• MapMode::Still will never complete if requests fail #1565
• Partial tiles constantly being reparsed #1458
• Ensure that LiveTileData can be reparsed #2393
• Hanging/crashing tests cutting-room-floor/tilelive-gl#25

These issues are often very difficult and time consuming to debug.

Yet despite its complexity, our current strategy is still inefficient:

• Don't continuously reparse tiles #2511

I'm not sure what the right approach is yet, however.

[tmpsantos]

We need a way of making tiles subscribe to the resources they need and get notified when they arrived. And I'm all in for getting rid of tile states as they are today. They will probably still exist, but IMO they could be inferred at runtime.

[kkaefer]

Partially addressed by #5139

[tmpsantos]

Partially addressed by #5139

👍

[jfirebaugh]

Lots of refactoring here lately. Fixing #667 will likely fix #2511 and then things will be pretty clean.