On November 28, 2021 12:01 PM, Eric Sunshine wrote: > On Sun, Nov 28, 2021 at 11:47 AM Baruch Burstein <bmburstein@xxxxxxxxx> > wrote: > > > When I perform these actions, I see a "fatal error" message which > > > clearly indicates failure, not success: > > > > > > % git worktree add ../foo bar > > > Preparing worktree (checking out 'bar') > > > fatal: 'bar' is already checked out at '.../wherever' > > > > I see them in the other order: > > > > fatal: 'master' is already checked out at 'C:/Users/bmbur/temp' > > Preparing worktree (checking out 'master') > > > > which I understood as "there was an error, but we managed to recover > > somehow and set up the worktree anyway" > > Okay, that's happening because the "Preparing" message is sent to stdout, > whereas the "fatal" error is sent to stderr, and the streams are being flushed > on Windows in a different order than what we typically see on Unix > platforms even though the "Preparing" message is actually output first by the > code. > > A general fix (not specific to git-worktree) might be to have die() (or maybe > vreportf()?) flush stdout before reporting the error message on stderr. That > should make output order more predictable, such that general status > messages appear before error messages. This is probably a good idea. There is no guarantee even in Unix regarding stream buffer flushing. If there is a reason to sequence output between stdout and stderr, then fflush is important. -Randall
