Thanks for the quick response.
The particular context in which I'm trying run this is in CLion. We've got various tools configured to run via `bazel run` (in particular, linting tools). CLion has a mechanism for executing external processes and dumping their outputs to an internal console. In that console, it attempts to color the text based on its stream. So, the normal usage arrives as red text, and error code arrives as white text. Not the end of the world. However, redirecting outputs in this pseudo-context doesn't work. So, it's a very limited, magical context. I'm trying to address this in multiple ways.
1. Get jetbrains to beef up their external tools.
2. Get options in bazel so I can suppress output I don't care about.
3. Simply wrap the thing in *another* executable and have *that* suppress the undesired outputs. I'm currently using this but I'd rather *not* have to increase the number of artifacts.