Mercurial Tip: Checking in regularly with uncommitted changes in your clone

WARNING: This blog entry was imported from my old blog on blogs.sun.com (which used different blogging software), so formatting and links may not be correct.

NetBeans, like OpenJDK and OpenSolaris, uses the Mercurial distributed version control system. I'm a big fan of of distributed version control. However, one thing that drives me nuts is this error message:

% hg merge
abort: outstanding uncommitted changes

This isn't just going to be a rant - I've finally found a solution which is working extremely well for me. I've suggested it to some other developers who have also reported that it works well for them, so I thought I would share it with you.

The reason I run into this all the time is my preferred style of work:

• I like to work on many things simultaneously. If while working on something, I come across a bug and I spot the problem, I just fix it right there and add a test for it. When I get ideas, I might go and put @todo tasks for myself, or leave editorial comments in various source files. I might also work on a couple of larger tasks simultaneously. Yes, I know the preferred Mercurial idiom for this is to have multiple clones, one for each task - but that isn't my preferred way of doing it. Each NetBeans clone is huge and takes a while to both clone and build - and I would have to switch my IDE editing context between these large source trees all the time.
  
  
• I like to check things in in logical chunks with a message that pertains to that particular change. This means I often check in a subset of the edited files in the tree.
  
  
• I like to check in code regularly rather than leaving all files edited until I'm ready to check everything in.
  
  
• I like to keep up to date. I'm using the bits I'm working on and I report problems when I see them so I want to make sure these are truly new and current bugs. Therefore, I update to the tip at least every day.
  
  

It turns out that the above requirements are tricky to fulfill with a default setup. Here's what happens.

• I'm working on something and a high priority bug comes in. I find and fix it. I check in just the fix for this one problem. I
  still have many other edited files in my clone since I was in the middle of something else.
  
  
  
• Since it's a high priority bug, I want to push it. However, when I do, I get this:
  
  

  
  % hg push
  pushing to https://tor:***@hg.netbeans.org/main/
  searching for changes
  abort: push creates new remote heads!
  (did you forget to merge? use push -f to force)
  

  
  
  This tells me that somebody has checked in something since my last pull. That happens a lot with a project like NetBeans
  where there are checkins every few minutes.
  
  
  
• No problem - I'll just pull down their changes, merge, and push again, right?
  

  
  % hg pull && hg update
  pulling from http://hg.netbeans.org/main/
  searching for changes
  adding changesets
  adding manifests
  adding file changes
  added 1 changesets with 1 changes to 1 files (+1 heads)
  (run 'hg heads' to see heads, 'hg merge' to merge)
  abort: crosses branches (use 'hg merge' or 'hg update -C')
  

  
  
  
• 
  Ah, I couldn't just update, I have to merge since I have checked in changesets of my own. Let's do that:
  

  
  % hg merge
  abort: outstanding uncommitted changes
  

  
  
  !@#@%*#%$%&*$ ! Now, if I have dozens of changes, what do I do? When I first started using Mercurial, I would either
  try to finish my other changes and check them in and then try the merge again, or if I really wasn't ready to do that,
  I would copy the contents of my modified files to another location, hg revert the changes, and try again.
  

It's not unreasonable of Mercurial to insist that you cannot merge with uncommitted changes. When it's merging, it will think
file modifications in the clone are your manual edits to resolve any merge conflicts. However, it does make using
my workflow difficult.




The Solution





It's really very simple: Use two clones. Do all your work in clone "main", and all your merging and pushing from clone "sync".
Let's say the main repository is "master", and your two clones are "main" and "sync":





To get set up, either clone the master repository twice, or, you can clone your main locally, but be sure to update
the "parent" reference in your sync clone such that it points to the master instead of your working clone
(I do that by copying main/.hg/hgrc to sync/.hg/hgrc.)






Here's the new workflow:




1. As before, work in main. Edit files at will, and check in logical changesets as needed.
   
   
2. When you want to push a changeset, go to your sync clone, and pull your new changesets from your working clone:
   

   
   % cd ../sync
   % hg pull ../main && hg update
   

   
   At this point, it's best to try a build/test as well, to make sure changeset is complete. If you're working
   on many things in parallel, it's possible that your changeset is depending on a change in a file you haven't
   checked in yet.
   
   
3. Now pull in the new changes from the master, and merge these:
   

   
   % hg fetch
   

   
   (Or if don't have the fetch extension installed, do it manually - hg pull && hg merge && hg ci -m "Merge").
   
   
4. If something went wrong with the merge - no problem. You can just go and nuke the entire sync clone
   and try again! Your modified files, and your new changesets, are still sitting completely unaffected in the main
   clone. Just clone sync again and try more carefully :)
   
   
5. Now you can push your merged changesets to the master repository:
   

   
   % hg push
   

   
   
6. ...and now you can pull these changes back into your main repository:
   

   
   % hg pull ../sync && hg update
   

   
   This will give you all the tip changes into your working clone, but without the risk of causing multiple heads that you
   have to merge. You've already merged your local changesets over in the sync clone, and therefore there is no
   conflict between your local changesets and the new changesets in the tip!
   
   


This process may seem tricky, but it's trivial once you try it:



Work in main, push and merge in sync, and pull back into main.




Finally, I want to call attention to item #4. Doing it this way means that it's trivial to try again if something wrong happens
during the merge. I've had a couple of merges where I've really mucked things up. Unfortunately, this was in my tree that contained
the changesets that I cared about. In the end I had to go and manually copy out the files I wanted to check in and try again. With
the above approach, if something goes wrong, just nuke the sync clone and try again.





This is the reason I'm suggesting this approach to anyone using Mercurial, not just people who want to work with edited files.
Especially when you're new to distributed version control systems or Mercurial, it's great to be able to go back if you make
a mistake. Just make sure you know what you're doing before you submit that final hg push command to push everything
back to the master repository!