Reviewing a Pull Request¶
An important step in our development workflow is the testing of individual issues/tickets after the development on them is complete, and before the code is merges into the main branch. Developers pick one from the list of completed issues and perform a number of verification steps on it. The mechanics of testing a pull request (e.g. the git commands to use) are described here. This page is concerned with the aspects that should be considered in deciding whether a pull request should be recommended to merge or sent back to the developer for further work. There should be very little reluctance to reopen a ticket even for minor issues.
The code changes should be manually reviewed (the github compare view is ideal for this). A couple of pieces on the value of code review can be found at scientopia and codinghorror.
The primary aim is to find bugs that the developer and tests so far have not spotted.
But also consider whether the code is ‘clean’, well-structured and easy to read/maintain.
Part of this is that:
There should be are no compiler (or doxygen) warnings coming from any modified classes
The code conforms to our coding standards.
Unit tests (or system tests if more appropriate) should be checked that they:
Exist and give adequate coverage (see unit testing practices).
If the ticket is fixing a bug there should be a test that makes sure we don’t have to fix the same bug again!
Do not load real data (data loading algorithms get a free pass on this one).
Leave the system in the same state that they found it (i.e. clean up).
Have a performance test, if appropriate.
Check that any user documentation is adequate and that there are release notes. In the case of new algorithms, there should be an accompanying
*.rstfile that has been added, containing an explanation of what exactly the algorithm does along with Python usage examples.
The first thing to note is that this should not just be a quick check of whatever the ticket says it does. Testing should be as much, if not more, about making sure the code does not do what it’s not supposed to do as that it does do what it’s supposed to.
All of the builds pass
The developer should have included instructions in the ticket of how to test things work.
But, as noted above, don’t just do that – also try to break it: click random buttons on GUIs, give unexpected/invalid inputs, etc.
Note down what you did in the ticket, and the platform you did it on.
If all the requirements have been met and documented approve the PR using GitHub’s review mechanism.
@mantidproject/gatekeepers group is who is meant to merge pull requests into master. This is done by social contract. A gatekeeper can
merge code if:
Green tick on the last build indicating all automated testing has succeeded
Adequate tests, both success and failure cases have been performed
There is comment on the code being reviewed