mozilla / multipreffer

Flippa buncha prefs

Multipreffer

• Allows defining a set of prefs per SHIELD branch, and sets them at install and cleans them up upon uninstall.
• Pref values are set on the default branch. User values are preserved.

The prefs to be set should be defined in src/variations.json, following this scheme:

{
  "branch 1": { // Name of branch
    "weight": 1, // Weight determines relative chance of getting assigned to this branch
    "prefs": { // Prefs and values to set upon install
      "pref1": "string1",
      "pref2": true,
      "pref3": 99
    }
  },

  [...]
}

[WIP] Process to develop a multipreffer-based study

TODO: streamline, automate.

1. Make a copy of the repo
2. Update metadata in src/manifest.json and package.json
3. Update abort and branch_name prefs
4. Define branches and target prefs/values in src/variations.json
5. npm install
6. npm run build
7. npm run test

If the tests pass, should be good to go! Build is in dist/ - upload to the study bug for signing.

multipreffer's Issues

Get "permanent" QA sign-off to reduce overhead for future experiments

This will involve:

• Determine and document a fixed process for building multipreffer experiments. (i.e. comprehensive instructions for developers and reviewers)
• Get QA sign-off for template
• Get sign-off from Normandy/Experiments team

Backport test fixes from etp search volume study

Backport the test fix in mozilla/etp-search-volume-study#3

Support a "post-experiment" set of prefs

Currently multipreffer caches the existing values of prefs before setting cohort values. But if a pref doesn't already exist, it's hard to know what to do at cleanup because default branch prefs can't be deleted. So it tries to set a false-y value depending on the type.

Maybe we can support a post-experiment cohort that lets the developer define what values to set at cleanup (these will last until a firefox restart)

Update readme to match new default-value version

Move repo to mozilla org

add prefhelper

in recent studies we've used "prefhelper" WEE to leave a breadcrumb with the experiment name and branch, as well as for an "aborted" pref. Should we bring this into the default build of multipreffert? And if we do, can we make the pref names auto-generated based off of the name in the manifest?

hide from about:addons

add "hidden": true to the manifest.json

Make a script to automate creation of multipreffer studies

Currently we need to create forks of the repo for every study. Instead, we should convert this repo into a "factory" rather than a template, and have a build script which accepts a config (addon name/descreption/id) and generates an add-on ready for signing.

This would eliminate the need for code review. Maybe we can extend variations.json to become the "config" file.

Make multipreffer import-able as a dependency in package.json

Make it easy to re-use multipreffer in other extensions that might want to also do their own thing on top of it

increase expiry to 365 days

We are using 365 days in all experiments based on this since the expiry within multipreffer does not actually work. This should be updated and noted here (until we manage to no longer depend on shield-utils).