Junio C Hamano <gitster@xxxxxxxxx> 于2021年4月22日周四 上午7:40写道： > > ZheNing Hu <adlternative@xxxxxxxxx> writes: > > > I admit that your idea makes sense, but we actually have another requirement: > > Construct a trailer with an empty value. > > It can be done with a different script given to .cmd, which would > say "exit 0" when allowing an empty string given as its input to > appear in the output. > Now I think that we should keep those trailers which ask for a "name <email>" pair, like "Helped-by", "Signed-off-by", when we provide a "help:","sign:" in command line, This allows the user to dynamically fill in the "name <email>" pair of other people in the commit message later. It is worthwhile for users to exit with exit(0). But those dispensable things like "Commit-Count", It must depend on a person's statistics in the git repository. So "cnt:" is meaningless, users' script can let it exit(1). > I was reacting what the "count" example does, and found that > counting commits by all authors (that is what an empty string would > match when given to --author="") somewhat illogical in the context > of that example. > The "Commit-Count" in the example here can only target a specific person, which has great limitations. I have a bold idea: Our current --trailer can only fill in one data item, and we don't expect it to fill in multiple rows. something like "Commit-Count", we hope to count the number of commits from everyone. But as we can see: Commit-count: 7 Linus Arver 1117 Linus Torvalds So If we have the opportunity to capture every line or every "block" of content, and feed it to git interpret-trailer, maybe we can get something like: Commit-count: 7 Linus Arver Commit-count: 1117 Linus Torvalds This will definitely make it easy for us to generate a lot of trailer at once. For example, a newbie like me, after making a patch, want to --cc all authors of one file, maybe I only need a command to get it. I don't know if it's a bit whimsical. > After all, these examples are to pique readers' interest by > demonstrating what the mechanism can do and how it can be used, and > for this feature, I think showing that > > (1) we can squelch the output from unasked-for execution; > > (2) we can reject --trailer=<key>:<value> when we do not like the > given <value>; > > (3) we can insert the trailer with the value we compute (and it is > OK for the computed result happens to be an empty string). > > should be plenty sufficient. OK. I will add these three examples in the new patch (when .cmd merge to master). Thanks. -- ZheNing Hu
