UX: Loading local directories

[ShrimpCryptid]

Description

• We want to allow users to try loading their own processed data as part of a tutorial/getting started guide WITHOUT having to host it via a cloud service (S3 etc.). Supporting loading local files and directories would make this possible and user-friendly (instead of users having to start a local server).
• I have a working prototype of directory loading available here: https://allen-cell-animated.github.io/timelapse-colorizer/pr-preview/pr-378

Screenshots from current prototype:

UX Problems to solve:

• Where should the new feature (loading directories) live in the existing loading dialog?
• Loading can take a VERY long time for files that are being served over a network. How do we indicate to users that a load is in progress?
  • This is not a problem for files that are actually stored/copied locally-- I just tested with the Nucmorph dataset and it loaded in a few seconds.
• Loading can't be cancelled once it starts without restarting the browser. (Also, only one load can happen at a time.) How do we indicate this to the user?
• If a user restarts the browser, we lose access to the original folder. What should they see on returning to the page?
• Should we indicate to a user that they're viewing a local file in the UI?
  • The code can tell the name of the directory that was loaded (but not the path). We could show it in the header...

---

Below to be filled out by UX during kickoff Q&A to the best of current understanding

Size

Effort level (xs, sm, m, lg, xl)

Scope

What will we do? What will we NOT do?

User needs statement(s)

As a [type of user], I need/want to [action(s)], so that I can [goal].

Additional notes

Knowns/unknowns, stakeholders, time/dependency considerations, deliverables

[toloudis]

More background / context:

1. We would like to have a tutorial that makes it easy for users to get their data into the viewer.
2. If the user will use our public viewer release to view the results of their work during the tutorial, then that data currently has to be stored in a location that can be served over https.
3. If the user was technically proficient enough to run the viewer locally, then they would be able to serve their own data over http (no "s") with a very simple one line python command.
4. We would love an even easier way so that users don't have to take extra steps to launch the viewer locally. If there were a VERY SIMPLE way to launch the viewer locally, then that might actually be the easiest.
5. An alternative, and the thing Peyton prototyped, is simply opening a local folder from the browser and loading data directly from it. Without using http.
6. The feasibility of this now leads to a whole new feature: LOAD DATA FROM A DIRECTORY THAT IS STORED LOCALLY.
7. We still have the problem of making an easy to use tutorial

[toloudis]

One alternative implementation - for the tutorial - could be to bundle a working version of the viewer as static files along with the tutorial so that a user could quickly load it without any special git or nodejs setup. The point of this alternative approach is that it would get us to a tutorial faster. It is independent of having the "load local data" as a new feature, which would still be very desirable and probably have a huge impact for users.

[ShrimpCryptid]

Additional notes from testing today:

• On supported browsers (Chrome, Edge, Opera, and other Chromium-based browsers) I can show a progress bar as the files are loaded. (Can track both total files and how many are currently loaded)
  • On unsupported browsers (Firefox, Safari) I can only show that a load operation is in progress, or when it finishes.
• In testing with large network datasets (nucmorph dataset) it took about 8 seconds after files were shown as "100% loaded" before the screen actually changed to the viewer.

[ShrimpCryptid]

Discussion:

• Could hide the feature (drag and drop only)
• Could expand a text area explaining the feature and how it's experimental, what to expect ("BETA/EXPERIMENTAL")
• Latency bug could be just my (Windows) setup, but no way to predict what other organizations have as their setups
• Shouldn't be in load modal?