Version

go 1.7 darwin/amd64

OS + Architecture

GOARCH="amd64"
GOEXE=""
GOHOSTARCH="amd64"
GOHOSTOS="darwin"
GOOS="darwin"
GORACE=""
GOROOT="/usr/local/Cellar/go/1.7/libexec"
GOTOOLDIR="/usr/local/Cellar/go/1.7/libexec/pkg/tool/darwin_amd64"
CC="clang"
GOGCCFLAGS="-fPIC -m64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -gno-record-gcc-switches -fno-common"
CXX="clang++"
CGO_ENABLED="1"

What did I do?

go --help and go [subcommand] --help

What did you expect to see?

I expected these commands to write to stdout. I was trying to see what the -i flag in go build -i does, so I ran the command go build --help | grep -- -i.

What did you see instead?

My buffer was flooded with the entire go build --help message because it writes to stderr instead of stdout. This means I have to run go {subcommand} --help 2>&1 | grep -- -{flag} when figuring out what a flag does.

However, when running commands like go env and go version, output is written to stdout. This leads to confusion: when should I have to combine stderr and stdout? What if there was a genuine error in the go {subcommand} that I run? I don't want to pipe that through grep and miss it, it may be important.

I think go should pick a file descriptor---either stdout or stderr---to write all go output to. In my opinion, go should only ever write to stderr if the binary itself fails in some respect. That way I don't miss any errors because I'm piping all of stderr through grep, head, and the like.