Traverse a used type and record all the dependencies we need to keep track of for incremental recompilation.
As a motivating example, given a type T
defined as:
type T >: L <: H type L <: A1 type H <: B1 class A1 extends A0 class B1 extends B0
We need to record a dependency on T
, L
, H
, A1
, B1
. This is
necessary because the API representation that ExtractAPI
produces for
T
just refers to the strings "L" and "H", it does not contain their API
representation. Therefore, the name hash of T
does not change if for
example the definition of L
changes.
We do not need to keep track of superclasses like A0
and B0
because
the API representation of a class (and therefore its name hash) already
contains all necessary information on superclasses.
A natural question to ask is: Since traversing all referenced types to
find all these names is costly, why not change the API representation
produced by ExtractAPI
to contain that information? This way the name
hash of T
would change if any of the types it depends on change, and we
would only need to record a dependency on T
. Unfortunately there is no
simple answer to the question "what does T depend on?" because it depends
on the prefix and ExtractAPI
does not compute types as seen from every
possible prefix, the documentation of ExtractAPI
explains why.
The tests in sbt types-in-used-names-a
, types-in-used-names-b
,
as-seen-from-a
and as-seen-from-b
rely on this.