|Version 12 (modified by benl, 5 years ago)|
- We use darcs for version control.
- For your first few patches, use darcs send to create a patch bundle, then email it to the disciple-cafe list.
- After those have been accepted, get an account on code.haskell.org. We'll give you commit access to the head repo and you can darcs push subsequent patches directly.
- Before sending or pushing patches do the following:
make cleanLibrary make cleanWar make war
This will clean out the libraries and test directory, rebuild the libraries, then run all the regression tests in the normal (non-optimised) way. This is the minimal required testing before pushing patches into the head branch. If you've made a deep change to something like the type inferencer or core transforms then you should run all the tests in all possible ways:make cleanLibrary make cleanWar make totalwar
If you've added a lot of files to the version control system (darcs add) and might have forgot some, or you just want to be extra sure you're not going to break anything, then it's best to use a separate testing repo. First make a new version of the head repo (maybe called ddc-head-test). Record your patches in the original repo, push them into the testing repo, then in that repo do:make cleantotal
Using this method has the additional advantage that in the testing repo you can leave the BUILDFLAVOUR set to distro. This means that the compiler itself will be built with optimisations turned on, and the tests will go through a lot faster.
- If your patch adds a new feature then it should also include a test case for it. Submit the test case in the same patch as the code for the feature.
- It's ok to push an "in progress" patch that cleans up existing code, or adds to it, without completing a feature. These patches don't need test cases, but the patch description should mention what it is working towards.
Handling Test Failures
- We don't like having failures in the main test suite.
- If the nightly buildbot encounters a failure when running a test in a non-standard way (like with "opt" or "prof") then someone should move that test into the test/Broken-skip directory and file a bug report.
- Don't push other, non-fixing patches into a repo that has test failures.
- Most of the bugs in the issue tracker should have a failing test case in test/Broken-skip
- When you fix a bug, mark it as resolved on the tracker then either move its test from test/Broken-skip into the main testsuite, or create a new test.
- If you find it too hard to create a new test, then we might need to extend the war test driver.