how to enforce consistent author and commit time stamps

[jghub]

I want to fix both author and commit timestamp. what I found is

JJ_TIMESTAMP $myts jj ci

which fixes the commit timestamp and

  jj metaedit --author-timestamp $myts -r @-

which fixes the author timestamp. however my tests seem to indicate that doing both in this order:

JJ_TIMESTAMP $myts jj ci
jj metaedit --author-timestamp $myts -r @-

does undo/overwrite the commit timestamp tweak, i.e. metaedit does reset it? is this the case or am I doing something wrong. also: what is the solution to get both timestamps set to the same time?

[jennings]

jj metaedit --update-author-timestamp should update both the author and committer timestamp to the current time. Or, I imagine JJ_TIMESTAMP=$myts jj metaedit --author-timestamp $myts would update them both to the specified time. Is that not what you're seeing?

The committer time is updated whenever a commit is written or rewritten, which explains why jj metaedit --author-timestmap also updates the committer time.

[jghub]

thanks. your last remark (metaedit will trigger commit timestamp update) was the relevant one: the only way (it seems??) to get both time stamps right is this order of actions:

jj metaedit --author-timestamp $myts -r
JJ_TIMESTAMP $myts jj ci

so that the commit-time overwrite triggered by metaedit happens first (might be obvious in retrospect but I did not see that...). is this somewhere in the help/documentation? I did not find any remark clarifying this situation (did not find JJ_TIMESTAMP at all -- only the internet knew...). I could imagine it is not a rare one: wanting to inject a sequence of commits (but not whole repo) from some other revision control system the pedestrian way while using the original timestamp(s).

[joyously]

maybe I do not quite understand your issue

Like you, I want to keep the correct timestamps of commits that I import. I already used your suggested command to change the first message, and got all descendants rebased with the same timestamp. But I don't want to change any of the messages, or any of the imported timestamps.

[jghub]

I am very new to jj so I do not know much. but it seems jj currently just does that: update all descandant time stamps in this situation. IIRC, git does only tolerate a (strictly?) monotonically increasing sequence of commit timestamps. so jj seemingly has to do that if the forged timestamp is newer than the oldest of its descandants. but AFAIU jj would not be forced to do that if that condition is not met. meaning: if first descandant is newer than the just forged timestamp of its parent jj could leave other timestamps alone. or so I presume. people in the know might correct if this is a misconception.

[joyously]

jj currently just does that: update all descandant time stamps in this situation
jj seemingly has to do that

Actually, I don't think that's correct (but I have not used Git much). I think it has more to do with the git hash includes the timestamp, and they point to each other by hash, so if you change one, you have to change the others (not sure if I'm confusing this with blockchain).

[jghub]

not a git user.... but I recall that Fossil when exporting a repo with fastexport for import into a git repo, the devs explicitly took care to remove what is called "time warps" in fossil speak: non-monotonic sections in the timestamp sequence (which are allowed in fossil). so there seems to be a constraint on the git side here. but can't say for sure.

[martinvonz]

IIRC, Git will use the timestamp as a heuristic for which side to walk in order to determine the common ancestor between two commits but I thought it would not affect correctness. If it walks the wrong side, it would just take longer until it realizes that it needs to walk the other side. Or something like that.

[jghub]

yes, you are probably right. thing is, the target repo in question (of course...) is a clone of a git remote and git log (what the others will use) shows author-timestamp while jj log (which is what I am trying/wanting to use) shows commit-timestamp. which is secondary but still somewhat confusing/irritating. but I agree, that commit-timestamp probably should better be left alone, usually.
but be that as it may, regarding

I think JJ_TIMESTAMP probably exists mainly for testing, which is why it's not documented well.

is it documented at all? I really did only find it in github-discussion and via gemini... also, there still might be valid use cases for this so it principally should be properly documented I believe

[jennings]

FYI, you can show author timestamps instead by redefining the template that displays timestamps:

[template-aliases]
'commit_timestamp(commit)' = 'commit.author().timestamp()'

[jghub]

that's nice, thank you. I think I will do that and then leave commit-timestamp alone