Message ID | 20250116012524.1557441-1-gitster@pobox.com (mailing list archive) |
---|---|
Headers | show |
Series | Send help text from "git cmd -h" to stdout | expand |
On Wed, Jan 15, 2025 at 05:25:17PM -0800, Junio C Hamano wrote: > Jonas Konrad noticed[*] that "git branch -h" started showing the help > text to the standard error stream. It turns out that we are fairly > inconsistent in our implementation of "git cmd -h". The users of > parse-options API will get "If -h is the only option on the command > line, give the help text to the standard output" for free, but some > commands manually check for the condition and then call the > usage_with_options() function, which gives the identical help text > to the standard error stream. And "git branch -h" Jonas noticed was > one of them. > > Older commands written before parse-options API became dominant show > the help text by calling the usage() function, which is meant to be > used when they fail to parse their command line arguments, which has > the same problem. An explicit request for help text "git cmd -h" > should be fulfilled by showing the help on the standard output. > > This series teachs "git $cmd -h" to send its help text to the > standard output stream consistently for built-in commands. I had a small complaint in patch 4, but otherwise this looks good to me. If we want to switch the exit code for this case from 129 to 0, I think we could easily do so on top (it would need modifications in three places, but now that you've untangled all of the individual builtins, that would get all of them). I guess there may be non-builtins that would need to be handled individually, though. We don't have too many of them these days, but they are not covered by t0012. -Peff
Jeff King <peff@peff.net> writes: > If we want to switch the exit code for this case from 129 to 0, I think > we could easily do so on top (it would need modifications in three > places, but now that you've untangled all of the individual builtins, > that would get all of them). Yes, I consider it pretty much an orthogonal issue to update the exit status. > I guess there may be non-builtins that would need to be handled > individually, though. We don't have too many of them these days, but > they are not covered by t0012. Yes. We can probably leave them as they are, as we have established our expectation with this series (i.e. an explicit end user request for help text should emit to the standard output stream), so any bug report can be handled without needing any policy discussion. "Hey, I noticed 'git cmd -h' writes to stderr, not stdout"---"Thanks, cmd need to be fixed".