tree a99b4a6010244aa282bdf828ba596bf4c5bf39db
parent 0de6f0fbeb0b0daae021c752ec1391d9d8d29dc6
author Nate Bosch <nbosch@google.com> 1605904226 -0800
committer GitHub <noreply@github.com> 1605904226 -0800
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJfuCdiCRBK7hj4Ov3rIwAAdHIIAJK0W37MITM9GehHYwF8Qo0g
 P3cidyymJEPtwiPJAPptQb5TW8XMDfFCcB3aKiU42vVJnGShWCUYonFsM0dnhiL/
 hgO78kUcizY7fT62PX4SXy6SSCLN3rzPjAtFdM7Oe4ts5ZEyxANUR5hepzpyy852
 SJgenCFC22ybNpXTBhcNSCJXb9lhSzo/xa5R5xsoS8QG+r6sn/707Yn6Fhx7NnsA
 0fO6LNV/uQIvCxvpiYZOdwWDCpZoeUDeuf+gj7QSIeoWdwZgpIeLsE6xqpBzEaGd
 1I2C/ytdqCZJpO3lVD0SsBHcPr/HloRJomvrAmEsh2Met0lPAxCVn8DbaQ0zj3Y=
 =Z29V
 -----END PGP SIGNATURE-----
 

Fix type errors for node socket channel (#1385)

The `StreamChannelTransformer` has tricky edge cases around subtyping
since it's generic type arguments should be contravariant instead of
covariant. Since there are different subtype relationships necessary
between the `sink` and `stream` transforms it is easier to work with by
splitting the transformation steps rather than modeling it exclusively
with multiple `StreamChannelTransformer` steps.