git: detect file creator

Hello,

I run a book digitizing project and pay people certain rate per 10K characters for the text files they upload to a git repo. Till now I was using following command to detect files authored by CertainEditor:

    git log --use-mailmap --no-merges --author="CertainEditor"
--name-only --pretty=format:""

Then I would pipe this output in `wc -m` and get amount of characters authored by CertainEditor and pay him accordingly. Usually editors do not touch each other's files and everything worked well. However recently one editor spotted a typo in somebody else's file and corrected it. This behavior is actually good and I would like to encourage it. However, now the command above lists the corrected file also as his, and so he gets paid for all the characters in the file while he changed only one of them. This, obviously, is not good.

1. Do you have an idea how can I list all the files **created** (not
authored / committed) by a user, so I can implement a fair characters
counting?

2. Maybe some commit hooks can be used that will check whether the
Author of a new commit is different from the previous one and if true
- override it to the previous Author?

3. Those small changes by a non-creator may be left not paid for (as
this action is not so intensive and may be reciprocal), but if you
have a good idea how I can pay for the "diff" the non-creator provides
- it would be nice! Do you think this "diff" should be deducted from
the creator? And if yes - how?

Thank you!

On Fri, Jul 15 2022, Sim Tov wrote:

It's nice to see that someone still believes in The Mythical Man- Month :)

If you want to adapt your current script perhaps --diff-filter helps, but...

..it seems you should fundamentally stop using it, and instead iterate over the commits, and pay for a "diff". Then you'd get the original change, as well as the change-on-top.

Just wc -l on the changed files(s) before & after, and pay the abs() difference.

You could walk it back with "git blame" I guess.

But you might want to consider the economic & social mis-incentives of lifting money from your co-workers coffers by pointing out a mistake to them...

Thank you very much, again, for the very useful suggestions!

I added `--diff-filter=AR` to my original command like this:

git log --use-mailmap --no-merges --diff-filter=AR --author="CertainEditor" --name-only --pretty=format:""

and it seems to do the job! May I have missed/messed something here?

2.1. Will such iteration over history not more time consuming compared to my command? 2.2. Will it not account for useless "diff"s, like add some rubbish, delete that rubbish - and I'll have to pay for it...

I pay per character, so you probably mean `wc -m`... but what do you mean by changed files? The command above is part of a script I run all the time I want to measure each editors' char count... it might be even without any specific recent changes...

Good point :-) I will not do it.

Now I see that if one editor renames(/moves) the files created by another editor - the former gets credits on all the characters inside those renamed files. This is bad. And it seemingly stems from the fact that in `--diff-filter=AR` means `A` OR `R`. Is there a way to do `A` OR (`A` AND `R`), meaning if a file was renamed then list it only if he also was created by that same author...

PS: the only way I see to approach it is to create a mechanism that
will prevent renaming
files by anyone except for the file creator (the one who Added file).
Can such a commit hook be created? If yes - how?

Sim Tov <smntov@gmail.com> writes:

As Git is distributed, local hooks cannot fundamentally be used as an enforcement mechanism.

How about running "git blame" on the end-result, with -C/-M turned on, instead of paying too much attention to file creation? That way, at least you can trace who contributed the words on each line.