Manual "new"

[spoerri]

On the back of #4012, it'd be nice to reduce the dependency on empty commits. Here's a different way to achieve that:

1. introduce a command jj view that loads the specified revision into the working copy, and makes all its files and directories readonly.
2. when abandoing a commit, delete that commit. if the commit is currently associated with the working copy, then use the same approach as jj view, i.e. make the working copy readonly.

[emilazy]

How can you make the working copy read‐only? Jujutsu can’t stop files from changing underneath it (unless you think we should require use of a VFS). Would it just throw an error and refuse to run if any of them differ?

[spoerri]

You could argue it’s just for GitHub interoperability and maybe for advance-branches.

[arxanas]

I don't think jj commit particularly helps with GitHub interoperability, at least not more than jj describe does? The restriction AIUI is just that you can't push commits without descriptions.

I don't see a reason to remove/deprecate jj commit; whom does that help? (Also, if it gets special advance-branches treatment, then it probably improves the case for not removing it.)

[spoerri]

Sorry, was abbreviating re: needing to care about @-, needing to have empty commits in order for things to work, and using jj commit to finish work: I have seen a sentiment that these are only a problem for using git forges.

For example, will certainly work with github if you do jj new; echo bar > baz; jj commit -m foo; jj git push -c @- and wishing that it would work with even jj new -m foo; echo bar > baz; jj push (notice the the actual change happens after the "describe").

It's unfortunate to need users to run both "new" before their changes and "commit" after their changes. IMHO eventually jj git push should just take care of it: update the branch from the working copy and push it.

[arxanas]

wishing that it would work with even jj new -m foo; echo bar > baz; jj push

I don't understand why this doesn't work now? It seems to work for me if I run jj new -m foo and then jj git push, because the current commit has a non-empty message.

[spoerri]

i guess it's really just the branches. would like them to automatically advance on jj git push