Running maintq try-work currently reports entries like:

maintq $ go run . try-work
waiting: <
  project: "pkgsite"
  branch: "master"
  change_id: "I02c38d1b90b10514f4bf50f21e6709e8829718dd"
  commit: "0651daffe465fa7b5a73fdb66884e9fe26fbcef6"
  version: 2
  go_commit: "e1b305af028544e00a22c905e68049c98c10a1cc"
  go_commit: "ef3039e99d3413735d863d84347561003d63340e"
  go_commit: "162b65e05cec843b1a6bee493362f2a3046dba7b"
  go_branch: "master"
  go_branch: "release-branch.go1.15"
  go_branch: "release-branch.go1.14"
  go_version: <
    major: 1
    minor: 15
  >
  go_version: <
    major: 1
    minor: 15
  >
  go_version: <
    major: 1
    minor: 14
  >
>

Note that the Go release branches "release-branch.go1.15" and "release-branch.go1.14" correspond to Go versions 1.15 and 1.14, which is accurate. However, the branch "master" (commit e1b305a) is reported to be Go version 1.15. It should be Go 1.16.

We're currently relying on a very simple heuristic for determining the Go version of the "master" branch, which is to take the version of the latest Go release. A much better heuristic is to report that branch (and any other non-release branch, like "dev.link") to be Go X.(Y+1).

There's room for improvement to precision here by reading the internal/goversion package in the git tree (issue #42376), but as a smaller step let's just make the current simple heuristic better. This is a piece needed to make a Go 1.16+ TryBot added in #42341 work.

CC @golang/release.