CL 166985 broke the build, but in a way that the build system didn't detect: it produced a build that couldn't be used to as a GOROOT_BOOTSTRAP directory.
We should add a test (perhaps off by default, except on at least 1 builders, including trybots) that just runs make.bash again using the just-built tree as its GOROOT_BOOTSTRAP.
Because we shard tests and slow tests are scheduled first, it shouldn't affect the overall trybot time.
/cc @mdempsky @bcmills @dmitshur
CL 166985 broke the build, but in a way that the build system didn't detect: it produced a build that couldn't be used to as a GOROOT_BOOTSTRAP directory.
We should add a test (perhaps off by default, except on at least 1 builders, including trybots) that just runs make.bash again using the just-built tree as its GOROOT_BOOTSTRAP.
Because we shard tests and slow tests are scheduled first, it shouldn't affect the overall trybot time.
/cc @mdempsky @bcmills @dmitshur