| Message ID | pull.892.git.1614793491538.gitgitgadget@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | branch: return error when --list finds no matches | expand |
"Josh Hunt via GitGitGadget" <gitgitgadget@gmail.com> writes: > From: Josh Hunt <johunt@akamai.com> > > Currently git branch --list foo always returns an exit status of 0 even > when the branch being searched for does not exist. Now an error is printed > and returns a non-zero exit status. Explaining what happens in the current code upfront is a good thing and is in line with the convention used in our project, which is good. But drop "currently" from there. Strictly speaking, it is not "always". In a corrupt repository, it is likely to show a proper error message and die. Also explaining what you want to happen before the end of the log message is good. But the proposed log message lacks why it is a good idea to make such a change, which is the most important part. If you ask me, I would say that the command was asked to show any branches, if exist, that match the given pattern, and did what it was asked to do without encountering any error---it just happened to have seen 0 branch that matched. So I think returning non-zero status would be a bug. Thanks.
On 3/3/21 5:27 PM, Junio C Hamano wrote: > "Josh Hunt via GitGitGadget" <gitgitgadget@gmail.com> writes: > >> From: Josh Hunt <johunt@akamai.com> >> >> Currently git branch --list foo always returns an exit status of 0 even >> when the branch being searched for does not exist. Now an error is printed >> and returns a non-zero exit status. > Junio Thanks so much for the review. > Explaining what happens in the current code upfront is a good thing > and is in line with the convention used in our project, which is > good. But drop "currently" from there. > > Strictly speaking, it is not "always". In a corrupt repository, it > is likely to show a proper error message and die. > > Also explaining what you want to happen before the end of the log > message is good. Thank you. I will make the above changes to the commit message if this moves forward. > > But the proposed log message lacks why it is a good idea to make > such a change, which is the most important part. OK sure. To provide better context I came across this when using 'git branch --list foo' in a script and there was an expectation (possibly incorrect) when the search was not successful it would return an non-zero exit status. I don't know if there's a convention that git follows, but I have the following examples where a similar type of command does return a non-zero exit status: johunt@johunt-ThinkPad-T480s:~$ ls foo ls: cannot access 'foo': No such file or directory johunt@johunt-ThinkPad-T480s:~$ echo $? 2 johunt@johunt-ThinkPad-T480s:~$ touch foo johunt@johunt-ThinkPad-T480s:~$ grep bar foo johunt@johunt-ThinkPad-T480s:~$ echo $? 1 johunt@johunt-ThinkPad-T480s:~$ grep . foo grep: foo: No such file or directory johunt@johunt-ThinkPad-T480s:~$ echo $? 2 Not just these tools, but even in git itself: johunt@johunt-ThinkPad-T480s:~$ git init foo Initialized empty Git repository in /home/johunt/foo/.git/ johunt@johunt-ThinkPad-T480s:~/foo$ touch bar johunt@johunt-ThinkPad-T480s:~/foo$ git add bar johunt@johunt-ThinkPad-T480s:~/foo$ git commit -am 'initial commit' [master (root-commit) 771510f] initial commit 1 file changed, 0 insertions(+), 0 deletions(-) create mode 100644 bar johunt@johunt-ThinkPad-T480s:~/foo$ git log foo fatal: ambiguous argument 'foo': unknown revision or path not in the working tree. Use '--' to separate paths from revisions, like this: 'git <command> [<revision>...] -- [<file>...]' johunt@johunt-ThinkPad-T480s:~/foo$ echo $? 128 johunt@johunt-ThinkPad-T480s:~/foo$ git grep foo . johunt@johunt-ThinkPad-T480s:~/foo$ echo $? 1 However it does seem like 'tag --list' follows the branch behavior: johunt@johunt-ThinkPad-T480s:~/foo$ git tag --list foo johunt@johunt-ThinkPad-T480s:~/foo$ echo $? 0 There are likely other examples of git commands showing both behaviors. Is it that branch and tag are a certain type of command which allows them to behave differently than say log or grep? > > If you ask me, I would say that the command was asked to show any > branches, if exist, that match the given pattern, and did what it > was asked to do without encountering any error---it just happened to > have seen 0 branch that matched. So I think returning non-zero > status would be a bug. Interesting. Going back to the scripting context then your suggestion would just be to check if output is NULL? This is what I've converted my scripts to do for now, but still feel like if --list can't find the branch I'm looking for then it should return non-zero as it matches behavior of other tools :) Josh
diff --git a/builtin/branch.c b/builtin/branch.c index bcc00bcf182d..5573175734fe 100644 --- a/builtin/branch.c +++ b/builtin/branch.c @@ -426,6 +426,8 @@ static void print_ref_list(struct ref_filter *filter, struct ref_sorting *sortin memset(&array, 0, sizeof(array)); filter_refs(&array, filter, filter->kind | FILTER_REFS_INCLUDE_BROKEN); + if (!array.nr) + die(_("no branches found")); if (filter->verbose) maxwidth = calc_maxwidth(&array, strlen(remote_prefix));