On Tue, 14 Sep 2021 13:25:16 +0200 Phil Sutter wrote: > Printing errors to stderr is useful to compare failing tests against an > expected set of failures - it is simply a task of comparing output on > stderr with a recorded one. I see. I'm still not sure the expected convenience factor (avoiding some grep-like post processing? but couldn't you compare the combined output just the same?) outweighs the weirdness / least surprise factor (having two variations of the same diagnostic message split on two separate output streams). > To not overcomplicate things, maybe the easiest fix would be to print > colors only if both stdout and stderr are a tty. What do you think? Yes, if the split has to stay, I don't have a better suggestion. Thanks, Štěpán
