FR: Config option to preserve timestamp on history rewrite

[jakobhellermann]

Problem

Sometimes I want to rewrite my commit history, without resetting every commit's committer timestamp to the current date.
Possible use cases include:

• purging a file accidentally committed in an early commit
• rewriting all commit messages into a common format before releasing a repo that was being developed in private before

You could argue that that's what the author timestamp is for, and that the commit timestamp should rightfully be changed in these instances, but the reality is that most code forges show the commit timestamp by default, and I don't think JJ should stand in the way of people wanting to control what the value of commit timestamps are, even if it doesn't entirely match reality.

EDIT: For some reason I thought I remembered that github and other forges show commit timestamp by default, not author timestamp. Maybe this is not be the case? I need to check. But the feature requests is still valid for author timestamps.

Possible Solutions

The obvious ideas for how to support this would be a command line flag or a config option, applying globally to every commit-rewriting operation.

Arguments in favor of config option:

• I'm assuming this is a niche use case and doesn't need to clutter up the valuable help text and tab autocomplete
• config options are more flexible, if you wanted to you could configure something like JJ_KEEP_COMMIT_TIMESTAMP=true jj sq -t foo to work
• with a config option I can set the option, do some jj edit -r rev and make some changes and unset the option when I'm done rewriting.

Arguments in favor of CLI flag:

• A config usually is configured persistently. I think this will be used more by one-off (or a couple) commands. This can be achieved by setting --config on the CLI so that's fine, but it feels a tiny bit like abusing configs.

I'd argue in favor of a config option.

Alternatives considered

Use something like git filter-repo in a colocated repository, and fix up the repo afterwards.

[PhilipMetzger]

It's very funny to see the counterpart to #2000 some years later. As someone who preferred the old behavior I'd like to see it happen.

[scott2000]

It's very funny to see the counterpart to #2000 some years later. As someone who preferred the old behavior I'd like to see it happen.

This seems independent from #2000? That issue was about updating the author timestamp in a specific case, whereas this seems to be about not updating the committer timestamp.

[dzaima]

For a feature of not updating the committer timestamp, a question would be what happens with the committer name/email if it's not equivalent to the current jj user.

If the name/email is updated, preserving the timestamp would be really weird behavior, mixing author timestamp and user from different sources. (but if you update the timestamp too, then you have questions what to do about descendant timestamps, given that you should really preserve monotonicity, e.g. for #9617)

If preserved, you'd be making fake commits attributed to other people (and screaming torvalds at you; esp. if this'd be a config that you can forget to unset).

On flag vs config - given that you should basically never want it to be permanently on (considering that it makes author and committer equivalent), I don't think it makes any sense as a config option.

On the actual feature - some form of general choice for "preserve/update committer/author timestamp/user" would be nice; would've helped for a script I wrote for importing git stashes where I ended up needing to do some annoying usage of the (undocumented) JJ_TIMESTAMP.

On flag clutter - imo showing the (full description of) global flags in all --help is already much much much too verbose, and a problem that is already in need of fixing, and there are already a couple flags that many will never need to use; I have for myself a shell function of jj --color=always "$@" --help | grep --color=auto -B9999 'Global Options:' for "jj --help but without the global flag spam". Autocomplete can't really be reduced, but there you have prefix matching anyway.