Before You Code
If you consider the following items before you begin coding, you will produce a patch that is easier to assess and more likely to be accepted:
- What problem are you trying to solve? Prior to any patch you should have a problem statement, and it should be included in the description of the patch.
- What is your solution approach? Please make this clear in the description, so that it does not have to be inferred from the code.
- Have you vetted your idea with the community? Please discuss on either the dev or clojure-dev list before your write your patch, and include a link to the discussion. (Not necessary for trivial patches like typos.)
- List the tradeoffs.
- What other approaches did you consider, what are their tradeoffs, and why is the approach you selected best?
- How will you prove to others your patch works? Plan to include tests. Example-based tests are ok, but generative tests are preferred.
- Don't do too much! Submit small patches that address specific problems, don't add anything extra, even (especially!) "cleanup" of nearby code.
While Your Patch is Being Considered
- Keep the description up-to-date as comments come in. It is very time-consuming to reconstruct the current state of a ticket by reading the comment thread.
- Rally support. Votes are good, and precise comments from other users reviewing the ticket are even better.
- As tickets grow, make sure you document clearly which patch(es) are active. Don't do this by deleting old patches, just refer to patches by name in the comments.
Once you're ready to craft your code, the first thing you'll need is a clone of the Clojure or appropriate repository. The examples below are for the Clojure project -- for submissions to Clojure itself:
$ git clone git://github.com/clojure/clojure.git
$ cd clojure
Next, create a new branch for yourself:
$ git checkout -b fixbug42
Switched to a new branch "fixbug42"
Now you're ready to get hacking. Make sure relevant doc strings are up to date and that all existing regression tests still pass e.g. for clojure core run:
$ ./antsetup.sh # Only needed once in a new Clojure tree, for ant to work
As noted in readme.txt, you will need to run ./antsetup.sh as a one-time setup before running ant. If you want to add new tests, that would be great too. Once you've finished making your changes you need to commit them.
$ git commit -a -m "fixed annoying bug, refs #42"
Created commit 8f7c712: fixed annoying bug, refs #42
1 files changed, 0 insertions(+), 1 deletions(-)
Now that you've made your changes it's time to get them into a patch. You need to update the repo and fix any conflicts you had.
$ git checkout master
Switched to branch "master"
$ git pull
$ git checkout fixbug42
Switched to branch "fixbug42"
$ git rebase master
Once you've fixed any conflicts, you're ready to create a patch:
$ git format-patch master --stdout > your-patch-file.diff
Now you can attach that patch file to the JIRA ticket. In the More Actions menu near the top of the page, select Attach Files. Please read and follow the recommendations below when writing comments about your attached patch. Screeners have limited time available for screening. You are more likely to get your patch approved if you can be as clear as you can, and as efficient with their time as possible.
- Include the file name and date of the patch in any comments referring to it. It is possible to match up comments with patches based on the date and time, but it is tedious and error prone.
- To get email whenever the ticket is updated, click on the word "Watch" in the top right area of the page. This can help you know when someone else comments on your patch, or creates a new one, etc. Click "Watching" if you want to stop these update emails for that ticket.
- If you create a new patch that incorporates one or more earlier ones, please combine them all into one patch file, and indicate in your comments that you have done this (with file names and dates of the patches you are superseding).
- If one of your patches becomes superseded by a later one, consider removing your patch to avoid confusion. See the instructions under the heading "Removing Patches" below.
Mark the ticket as having a patch ready for screening by editing the Patch attribute. Click the Edit button near the top left of the page for the ticket. In the next page find the heading "Patch" with a popup menu next to it. Select "Code" or "Code and Test" from that menu, then click the Update button at the bottom of the page. If you do not see an Edit button on the page for the ticket, and you have signed a CA, ask on the developer's email list to be given permission to edit Jira tickets.
To remove a patch (e.g. because it is obsolete), go to the page for the ticket and look for the "Attachments" heading beneath the Description text. Far to the right is a plus sign and a triangle. Click on the triangle and select "Manage Attachments" from the menu. Think carefully on which one you want to delete, and click the trash can icon next to it. Note: Most (or all?) people have permission to remove their own attachments, but not those added by someone else.
Updating stale patches
A stale patch means one that used to apply cleanly to the latest Clojure master version, but due to commits made since the patch was created, it no longer does. In particular, the output of this command:
% git am --keep-cr -s < patch_file
includes 'Patch failed' and 'To restore the original branch and stop patching, run "git am --abort"'. You should do the "git am --abort" to get rid of state of the failed patch attempt left behind by the command above.
"git am" is very "fragile", meaning that if the patch_file was created with one version of the source code, all it takes for the command to fail is a change in any of the lines of context present in the patch file, even if it is not one of the lines being changed by the patch. This is especially common for files containing unit tests, because people usually add new tests at the end of such a file, and so the lines of context before the new test change if two different patches add a new test to the end of the same file.
To apply such a patch, the 'patch' program by Larry Wall is extremely useful. It comes preinstalled with Mac OS X and most Linux distributions. You can easily install it with Cygwin for Windows.
% patch -p1 < patch_file
The output will give you some hints of whether each "hunk" of the patch file succeeded or failed. If they all succeed, then likely the only thing wrong with the patch file is that a few context lines were changed. If any hunks fail, patch creates files ending with ".rej" containing rejected hunks that it did not apply, and you can focus on those as places where the source code likely changed more significantly. A command like this will find them all:
% find . -name '*.rej'
You will need to look at those rejected hunks, perhaps think about them for a bit to see if and how they still apply, and apply them by hand-editing the source code yourself.
When recreating a new git format patch with:
% git format-patch master --stdout > patch_file
it puts your name and the current date near the top of the file. If the only changes that you have made are in the context lines, please keep the original author's credit intact by copying the name and date from the original patch that you started from, then upload that.
If you write unit tests where there were none in the original patch, but didn't otherwise modify the original patch, and you would like your name in the commit log for your work, create a separate patch of test additions with your name on it, leaving the original author's name on the updated patch.