Settings0 used to be a Map[Scope, AttributeMap], and is now a
Map[ScopedKey[x], x].
This is better because we don't need to decompose all ScopedKey[x]
into a Scope and an AttributeKey[x], for recomposing it back later,
which duplicates all ScopedKey[x]. It reduces the number of long-living
ScopedKey[x] by 8%, and the total number of instances by 1.4%.
Also it improves the performance of Settings0, which was responsible of
2.95% of the total CPU time, and is now responsible of 0.41%.
For the details about this PR, please see the blog post https://eed3si9n.com/sbt-remote-cache/.
* Add cache basics
* Refactor Attributed to use StringAttributeMap, which is Map[StringAttributeKey, String]
* Implement disk cache
* Rename Package to Pkg
* Virtualize packageBin
* Use HashedVirtualFileRef for packageBin
* Virtualize compile task
Problem
-------
In sbt 1, platform cross building is implemented using in the user-land
using `%%%` operator, which clevery handles both Scala cross building
and appending platform suffix like sjs1.
However, in general symbolic `%%%` is confusing, and hard to explain.
Solution
--------
In sbt 2, we should subsume the idea of platform cross building,
so `%%` can act as the current `%%%` operator.
This adds a new setting called `platform`, which defaults to
`Platform.jvm` by default.
When a subprojects sets it to `Platform.sjs1`, `ModuleID`s defined using
`%%` operator will inject the platform suffix `_sjs1` **in addition**
to the Scala binary suffix `_2.13` etc.
Note: Explicit JVM dependencies will now require `.platform(Platform.jvm)`.