On Tue, Nov 20, 2018 at 05:58:25AM -0500, Jeff King wrote: > On Tue, Nov 20, 2018 at 05:45:28AM -0500, Jeff King wrote: > > > On Tue, Nov 20, 2018 at 12:34:04AM +0100, SZEDER Gábor wrote: > > > > > > I do notice that many of the existing "FATAL:" errors use descriptor 5, > > > > which goes to stdout. I'm not sure if those should actually be going to > > > > stderr (or if there's some TAP significance to those lines). > > > > > > I chose to send these messages to standard error, because they are, > > > well, errors. TAP only cares about stdout, but by aborting the test > > > script in BUG(), error() or die() we are already violating TAP anyway, > > > so I don't think it matters whether we send "bug in test script" or > > > "FATAL" errors to stdout or stderr. > > > > Yeah, I agree it probably doesn't matter. I was mostly suggesting to > > make the existing ">&5" into ">&7" for consistency. But I don't think > > that needs to block your patch. > > By the way, I did check to see whether this might help the situation > where running under "prove" we see only "Dubious..." and not the actual > error() message produced by the test script. But no, it eats both stdout > and stderr. You can sneak a line through by prepending it with "#", but > only if "prove -o" is used (and even then, it's hard to associate it > with a particular test when you're running many in parallel).
Just to be clear: I don't mind if in some combination of test harnesses and test options a "bug in the test script" message doesn't reach the terminal as long as I get a clearly visible error from somewhere. > So I guess the status quo is not too bad: prove says "dubious", and then > you re-run the test with "./t1234-foo.sh -v -i" and you get to see the > error. And with '--verbose-log' the "bug in the test script" message goes to the test's log as well, even when it has to go through fd 7 first, so if you use 'prove' and your GIT_TEST_OPTS includes '--verbose-log' then you can just look at the log, there's no need to re-run the test.
