Contributing code to Scikit-beam, a worked example

Warning

This still needs to be adapted for skbeam

This example is based on fixing Issue 1761 from the list of Scikit-beam issues on GitHub. It resulted in pull request 1917.

The issue title was “len() does not work for coordinates” with description “It would be nice to be able to use len on coordinate arrays to know how many coordinates are present.”

This particular example was chosen because it was tagged as easy in GitHub; seemed like the best place to start out!

Short on time? Don’t want to read a long tutorial?

There is a minimalist, command-only version of this at Condensed example of contributing code to Scikit-beam. You should have pull request 1917 open as you read the commands so you can see the edits made to the code.

There is also a very exciting terminal-cast at Terminal-cast of worked example.

Before you begin

Make sure you have a local copy of scikit-beam set up as described in Try the development version. In a nutshell, the output of git remote -v, run in the directory where your local of Scikit-beam resides, should be something like this:

scikit-beam   git://github.com/scikit-beam/scikit-beam.git (fetch)
scikit-beam   git://github.com/scikit-beam/scikit-beam.git (push)
your-user-name     git@github.com:your-user-name/scikit-beam.git (fetch)
your-user-name     git@github.com:your-user-name/scikit-beam.git (push)

The precise form of the URLs for your-user-name depends on the authentication method you set up with GitHub.

The important point is that scikit-beam should point to the official Scikit-beam repo and your-user-name should point to your copy of Scikit-beam on GitHub.

Grab the latest updates to scikit-beam

A few steps in this tutorial take only a single command. They are broken out separately to outline the process in words as well as code.

Inform your local copy of Scikit-beam about the latest changes in the development version with:

git fetch scikit-beam

Set up an isolated workspace

  • Make a new git branch for fixing this issue, check it out, and let my GitHub account know about this branch:

    git branch fix-1761 scikit-beam/master   # branch based on latest from GitHub
git checkout fix-1761                   # switch to this branch
git push --set-upstream origin fix-1761 # tell my github acct about it

  • Make a python environment just for this fix and switch to that environment. The example below shows the necessary steps in the Anaconda python distribution:

    conda create -n apy-1761 --clone root  # Anaconda distribution only
source activate apy-1761

    If you are using a different distribution, see Python virtual environments for instructions for creating and activating a new environment.

  • Install our branch in this environment (will not work in python 3) with:

    python setup.py develop

Do you really have to set up a separate python environment for each fix? No, but you definitely want to have a python environment for your work on code contributions. Making new environments is fast, doesn’t take much space and provide a way to keep your work organized.

Tests first, please

It would be hard to overstate the importance of testing to Scikit-beam. Tests are what gives you confidence that new code does what it should and that it doesn’t break old code.

You should at least run the relevant tests before you make any changes to make sure that your python environment is set up properly.

The first challenge is figuring out where to look for relevant tests. Issue 1761 is a problem in the coordinates package, so the tests for it are in scikit-beam/coordinates/tests. The rest of Scikit-beam has a similar layout, described at Testing Guidelines.

Change to that directory and run the current tests with:

cd scikit-beam/coordinates/test
py.test

The tests all pass, so I need to write a new test to expose this bug.

There are several files with tests in them, though:

$ ls
test_angles.py
test_angular_separation.py
test_api.py
test_arrays.py
test_distance.py
test_formatting.py
test_matching.py
test_name_resolve.py
test_transformations.py

Issue 1761 affects arrays of coordinates, so it seems sensible to put the new test in test_arrays.py. As with all of the steps, if you are not sure, ask on the scikit-beam-dev mailing list.

The goal at this point may be a little counter-intuitive: write a test that we know will fail with the current code. This test allows Scikit-beam to check, in an automated way, whether our fix actually works and to make sure future changes to code do not break our fix.

Looking over the existing code in test_arrays.py, each test is a function whose name starts with test_; the last test in the file is test_array_indexing so an appropriate place to add the test is right after that.

Give the test a reasonably clear name; I chose: test_array_len. The easiest way to figure out what you need to import and how to set up the test is to look at other tests. The full test is in the traceback below and in pull request 1917

Write the test, then see if it works as expected–remember, in this case we expect to fail. Running py.test test_arrays.py gives the expected result; an excerpt from the output is:

================= FAILURES =============================
______________ test_array_len __________________________

    def test_array_len():
        from .. import ICRS

        input_length = 5
        ra = np.linspace(0, 360, input_length)
        dec = np.linspace(0, 90, input_length)

        c = ICRS(ra, dec, unit=(u.degree, u.degree))

>       assert len(c) == input_length
E       TypeError: object of type 'ICRS' has no len()

test_arrays.py:291: TypeError

Success!

Add this test to your local git repo

Keep git commits small and focused on one logical piece at a time. The test we just wrote is one logical change, so we will commit it. You could, if you prefer, wait and commit this test along with your fix.

For this tutorial I’ll commit the test separately. If you aren’t sure what to do, ask on scikit-beam-dev mailing list.

Check what was changed

We can see what has changed with git status:

$ git status
On branch fix-1761
Your branch is up-to-date with 'origin/fix-1761'.

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git checkout -- <file>..." to discard changes in working directory)

    modified:   test_arrays.py

no changes added to commit (use "git add" and/or "git commit -a")

There are two bits of information here:

  • one file changed, test_arrays.py

  • We have not added our changes to git yet, so it is listed under Changes not staged for commit.

For more extensive changes it can be useful to use git diff to see what changes have been made:

$ git diff
diff --git a/scikit-beam/coordinates/tests/test_arrays.py b/scikit-beam/coordinates/test
index 2785b59..7eecfbb 100644
--- a/scikit-beam/coordinates/tests/test_arrays.py
+++ b/scikit-beam/coordinates/tests/test_arrays.py
@@ -278,3 +278,14 @@ def test_array_indexing():
     assert c2.equinox == c1.equinox
     assert c3.equinox == c1.equinox
     assert c4.equinox == c1.equinox
+
+def test_array_len():
+    from .. import ICRS
+
+    input_length = 5
+    ra = np.linspace(0, 360, input_length)
+    dec = np.linspace(0, 90, input_length)
+
+    c = ICRS(ra, dec, unit=(u.degree, u.degree))
+
+    assert len(c) == input_length

A graphical interface to git makes keeping track of these sorts of changes even easier; see Get a git GUI (optional) if you are interested.

Stage the change

git requires you to add changes in two steps:

  • stage the change with git add test_arrays.py; this adds the file to the list of items that will be added to the repo when you are ready to commit.

  • commit the change with git commit; this actually adds the changes to your repo.

These can be combined into one step; the advantage of doing it in two steps is that it is easier to undo staging than committing. As we will see later, git status even tells you how to do it.

Staging can be very handy if you are making changes in a couple of different places that you want to commit at the same time. Make your first changes, stage it, then make your second change and stage that. Once everything is staged, commit the changes as one commit.

In this case, first stage the change:

git add test_arrays.py

You get no notice at the command line that anything has changed, but git status will let you know:

$ git status
On branch fix-1761
Your branch is up-to-date with 'origin/fix-1761'.

Changes to be committed:
  (use "git reset HEAD <file>..." to unstage)

    modified:   test_arrays.py

Note that git helpfully includes the command necessary to unstage the change if you want to.

Commit your change

I prefer to make commits frequently, so I’ll commit the test without the fix:

$ git commit -m'Add test for array coordinate length (issue #1761)'
[fix-1761 dd4ef8c] Add test for array coordinate length (issue #1761)
 1 file changed, 11 insertions(+)

Commit messages should be short and descriptive. Including the GitHub issue number allows GitHub to automatically create links to the relevant issue.

Use git status to get a recap of where we are so far:

$ git status
On branch fix-1761
Your branch is ahead of 'origin/fix-1761' by 1 commit.
  (use "git push" to publish your local commits)

nothing to commit, working directory clean

In other words, we have made a change to our local copy of scikit-beam but we have not pushed (transferred) that change to our GitHub account.

Fix the issue

Write the code

Now that we have a test written, we’ll fix the issue. A full discussion of the fix is beyond the scope of this tutorial, but the fix is to add a __len__ method to scikit-beam.coordinates.SphericalCoordinatesBase in coordsystems.py. All of the spherical coordinate systems inherit from this base class and it is this base class that implements the __getitem__ method that allows indexing of coordinate arrays.

See pull request 1917 to view the changes to the code.

Test your change

There are a few levels at which you want to test:

  • Does this code change make the test we wrote succeed now? Check by running py.test tests/test_arrays.py in the coordinates directory. In this case, yes!

  • Do the rest of the coordinate tests still pass? Check by running py.test in the coordinates directory. In this case, yes–we have not broken anything!

  • Do all of the scikit-beam tests still succeed? Check by moving to the top level directory (the one that contains setup.py) and run python setup.py test. This may take several minutes depending on the speed of your system. Success again!

Note

Tests that are skipped or xfailed are fine. A fail or an error is not fine. If you get stuck, ask on scikit-beam-dev mailing list for help!

Stage and commit your change

Add the file to your git repo in two steps: stage, then commit.

To make this a little different than the commit we did above, make sure you are still in the top level directory and check the git status:

$ git status
On branch fix-1761
Your branch is ahead of 'origin/fix-1761' by 1 commit.
  (use "git push" to publish your local commits)

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git checkout -- <file>..." to discard changes in working directory)

    modified:   scikit-beam/coordinates/coordsystems.py

no changes added to commit (use "git add" and/or "git commit -a")

Note that git knows what has changed no matter what directory you are in (as long as you are in one of the directories in the repo, that is).

Stage the change with:

git add scikit-beam/coordinates/coordsystems.py

For this commit it is helpful to use a multi-line commit message that will automatically close the issue on GitHub when this change is accepted. The snippet below accomplishes that in bash (and similar shells):

$ git commit -m"
> Add len() to coordinates
>
> Closes #1761"
[fix-1761 f196771] Add len() to coordinates
 1 file changed, 4 insertions(+)

If this was not a tutorial I would write the commit message in a git gui or run git commit without a message and git would put me in an editor.

However you do it, the message after committing should look like this:

Add len() to coordinates

Closes #1761

You can check the commit messages by running git log. If the commit message doesn’t look right, ask about fixing it at scikit-beam-dev mailing list.

Push your changes to your GitHub fork of scikit-beam

This one is easy: git push

This copies the changes made on your computer to your copy of Scikit-beam on GitHub. At this point none of the Scikit-beam maintainers know anything about your change.

We’ll take care of that in a moment with a “pull request”, but first…

Stop and think: any more tests or other changes?

It never hurts to pause at this point and review whether your proposed changes are complete. In this case I realized there were some tests I could have included but didn’t:

  • What happens when len() is called on a coordinate that is not an array?

  • Does len() work when the coordinate is an array with one entry?

Both of these are mentioned in the pull request so it doesn’t hurt to check them. In this case they also provide an opportunity to illustrate a feature of the py.test framework.

I’ll move back to the directory containing the tests with cd scikit-beam/coordinates/tests to make it a bit easier to run just the test I want.

The second case is easier, so I’ll handle that one first following the cycle we used above:

  • Make the change in test_arrays.py

  • Test the change

The test passed; rather than committing this one change I’ll also implement the check for the scalar case.

One could imagine two different desirable outcomes here:

  • len(scalar_coordinate) behaves just like len(scalar_angle), raising a TypeError for a scalar coordinate.

  • len(scalar_coordinate) returns 1 since there is one coordinate.

If you encounter a case like this and are not sure what to do, ask. The best place to ask is in GitHub on the page for the issue you are fixing.

Alternatively, make a choice and be clear in your pull request on GitHub what you chose and why; instructions for that are below.

Testing for an expected error

In this case I opted for raising a TypeError, because the user needs to know that the coordinate they created is not going to behave like an array of one coordinate if they try to index it later on. It also provides an opportunity to demonstrate a test when the desired result is an error.

The py.test framework makes testing for an exception relatively easy; you put the code you expect to fail in a with block:

with pytest.raises(TypeError):
    c = ICRS(0, 0, unit=(u.degree, u.degree))
    len(c)

I added this to test_array_len in test_arrays.py and re-ran the test to make sure it works as desired.

Aside: Python lesson–let others do your work

The actual fix to this issue was very, very short. In coordsystems.py two lines were added:

def __len__(self):
    return len(self.lonangle)

lonangle contains the Angle``s that represent longitude (sometimes this is an RA, sometimes a longitude). By simply calling ``len() on one of the angles in the array you get, for free, whatever behavior has been defined in the Angle class for handling the case of a scalar.

Adding an explicit check for the case of a scalar here would have the very big downside of having two things that need to be kept in sync: handling of scalars in Angle and in coordinates.

Commit any additional changes

Follow the cycle you saw above:

  • Check that all Scikit-beam tests still pass; see Test your change

  • git status to see what needs to be staged and committed

  • git add to stage the changes

  • git commit to commit the changes

The git commands, without their output, are:

git status
git add scikit-beam/coordinates/tests/test_arrays.py
git commit -m"Add tests of len() for scalar coordinate and length 1 coordinate"

Edit the changelog

Keeping the list of changes up to date is nearly impossible unless each contributor makes the appropriate updates as they propose changes.

Changes are in the file CHANGES.rst in the top-level directory (the directory where setup.py is). Put the change under the list that matches the milestone (aka release) that is set for the issue in GitHub. If you are proposing a new feature in a pull request you may need to wait on this change until the pull request is discussed.

This issue was tagged for 0.3.1, as shown in the image below, so the changelog entry went there.

../../_images/milestone.png

The entry in CHANGES.rst should summarize was you did and include the issue number. For writing changelog entries you don’t need to know much about the markup language being used (though you can read as much as you want about it at the Sphinx primer); look at other entries and imitate.

For this issue the entry was the line that starts - Implemented:

- ``scikit-beam.coordinates``

  - Implemented `len()` for coordinate objects. [#1761]

Starting the line with a - makes a bulleted list item, indenting it makes it a sublist of scikit-beam.coordinates and putting len() in single backticks makes that text render in a typewriter font.

Commit your changes to the CHANGES.rst

You can use git status as above or jump right to staging and committing:

git add CHANGES.rst
git commit -m"Add changelog entry for 1761"

Push your changes to GitHub

One last push to GitHub with these changes before asking for the changes to be reviewed:

git push

Ask for your changes to be merged with a pull request

This stage requires to go to your GitHub account and navigate to your copy of scikit-beam; the url will be something like https://github.com/your-user-name/scikit-beam.

Once there, select the branch that contains your fix from the branches dropdown:

../../_images/worked_example_switch_branch.png

After selecting the correct branch click on the “Pull Request” button, like that in the image below:

../../_images/pull_button.png

Name your pull request something sensible. Include the issue number with a leading # in the description of the pull request so that a link is created to the original issue.

Please see pull request 1917 for the pull request from this example.

Revise and push as necessary

You may be asked to make changes in the discussion of the pull request. Make those changes in your local copy, commit them to your local repo and push them to GitHub. GitHub will automatically update your pull request.