tree 1f14c99c5d4bbe0feb6b6e453fd1f36a20a21b90
parent 7a6d7f4c217005c7a6b9b7e6f38b00258c530b11
author Joshua Trask <joshtrask@google.com> 1635801177 -0400
committer Joshua Trask <joshtrask@google.com> 1636408623 -0500

Preliminary "delegate chooser" unbundling.

Migrate the "delegate chooser" activity from SysUI to the
new Sharesheet (IntentResolver) package. Set up basic
project structure, tests, etc.

This CL involves a lot of unfamiliar boilerplate that I've
tried to copy from neighboring projects. Please feel free to
recommend additional reviewers for any complexities I may
have overlooked.

The rest of this CL description will detail:

  1. Unbundling/"delegate chooser" background
  2. Scope: what's included in this change
  3. Scope: what's not included in this change
  4. Key considerations for code reviewers

[1. Unbundling/"delegate chooser" background]

The "delegate chooser" activity will eventually replace
the system `ChooserActivity` as the new "unbundled"
Sharesheet implementation. In order to validate the new
architecture, an earlier exploration added a "trampoline"
chooser implementation in SysUI. This implementation
presents no UI of its own, but instead handles the
fulfillment of the user's selection from the system
Sharesheet. In upcoming experiments, we expect to have
the unbundled implementation replicate the original UI of
the system `ChooserActivity` so that the new version can
take over earlier in the flow, and ultimately replace the
system implementation altogether. For more information,
see go/sharesheet-unbundling.

[2. Scope: what's included in this change]

This CL replicates SysUI's "trampoline" delegate
implementation into the newly-created directory for the
unbundled module. The main purpose of this CL is to
set up the basic boilerplate in that directory (owners,
etc). I stopped at the first meaningful milestone that I
could confirm worked correctly -- the unbundled unit
tests pass with atest. Note that (other than the package
names) the Java code itself is unchanged from the
earlier SysUI implementation.

[3. Scope: what's not included in this change]

Before the longer-term improvements outlined in
go/sharesheet-unbundling, there are a few fundamental
unbundling steps that are remaining after this CL. They
were omitted in this first pass since we already have a
working milestone and intend to focus the current review
on project/directory boilerplate.

Specifically:

 A. There is no experiment that would cause users to
    exercise the new delegation path (b/202166034).
    If there are any issues with the implementation at this
    point, we wouldn't expect them to be visible to users.

 B. There may(?) be other steps required to integrate the
    new "unbundled" module into the Android build (e.g.
    makepush?). This CL is sufficient to run the unit tests
    in the new unbundled directory, but no manual testing
    was done to confirm the end-user behavior.

 C. This doesn't include any APEX boilerplate (manifests,
    keys). That may not be important yet since we don't
    have any immediate plans to launch as a mainline
    module, or it may come up as part of doing the
    integration work for (B).

 D. The old SysUI delegate implementation is left in tact,
    but should be cleaned up as we switch over to the new
    architecture.

[4. Key considerations for code reviewers]

This change is mostly boilerplate cribbed from neighboring
projects. Following is a short list of decisions I made
that could deserve a second look:

 A. The package name: com.android.intentresolver

 B. (omitted for lint, but this was resolved in a
     discussion with OSPO)

 C. The versioning scheme. I used a YYYY-MM versionName
    as in ExtServices, but for now just have
    versionCode="0" until we have a numeric scheme.

 D. AndroidManifest attributes, mostly copied from SysUI.

 E. SDK versions / "platform_apis" attribute. I had to try
    a few things to get this working. It seems OK now, but
    there may be a better/more elegant approach. If we do
    need to support specific earlier versions, we should
    confirm that those versions had the
    startActivityAsCaller API available. Either way,
    there's some cruft in this CL that should be cleaned
    up once we confirm the approach.

Bug: 202164690, 202165476
Test: `atest IntentResolverUnitTests`
Change-Id: If53eee98ac92685810bbb84feb9cb1030a161502