Warn/require confirmation on bookmarking empty changes?

[sourcefrog]

In getting to know jj, when I am working on a feature branch that I want to push to github, I've fairly often run a flow like

jj commit -m Whatever
jj bookmark move featurebranch

The trap of course is that the bookmark is set to the new empty revision following the thing that I committed. I suppose pushing the empty rev to a feature branch on github is not too harmful but it does seem a bit messy.

The right answer I guess is that I should either

1. Move the bookmark before committing, which seems a bit weird coming from other VCS
2. bookmark move -r@-
3. Just jj desc then bookmark it but then don't forget to jj new

I see there is a good bit of existing discussion about various types of auto-moving bookmarks, which would perhaps be a more fundamental answer for this kind of workflow:

• Feedback for the experimental `advance-branches` feature #3549
• Working branches and the JJ "way" #2425
• How bookmarks and Git ought to interact #5333
• FR: Topics (alternative to branches) #3402
• FR: --advance-branches flag for jj commit #2338

As a more concrete and tactical (perhaps excessively tactical) fix I wonder if the UI should by default prevent moving a bookmark to an empty change, the same way it by default prevents you pushing new bookmarks and moving bookmarks backwards.

That is, jj new; jj bookmark set bbb would fail unless you said --allow-empty.

Alternatively, rather than failing it could say

warning: bookmark bbb is set to an empty revision
hint: to set it to the parent revision use `jj bookmark move bbb -rrstuvwxyz --allow-backwards`

[yuja]

Doesn't jj git push fail if commit had no description?

I don't think jj bookmark should fail depending on the commit content because it's totally fine to bookmark an empty commit (including merges.) To prevent common mistakes, it might be better to require explicit revision argument to bookmark (i.e. remove the default @.)

[sourcefrog]

Doesn't jj git push fail if commit had no description?

It does, but then you have the slightly awkward situation where you have to realize to move the bookmark backwards.

I don't think jj bookmark should fail depending on the commit content because it's totally fine to bookmark an empty commit (including merges.)

Oh, I see what you mean about merges. But wouldn't bookmarking an empty single-parent empty-metadata still be rare?

To prevent common mistakes, it might be better to require explicit revision argument to bookmark (i.e. remove the default @.)

That might be another good option.

[drieber]

I filed #5374 for this and I am working on a fix.