Mercurial test suite is a combination of shell scripts with reference outputs and a few python scripts, unusable on a pure Windows platform. pysh is a unix shell implementation in python designed to run as many tests as possible. While running almost 90% of the test suite, the implementation suffers of the following limitations:
- Many utilities are still to be implemented or replaced, which explains why Cygwin is required.
- Program outputs like filenames and system messages are hard to simulate. Current solution is to patch the tests to rewrite the output as expected. One patch queue is maintained on top of crew repository.
- Asynchronous lists are supported, but kill sometimes refuses to kill and the test is stuck waiting for a process to complete. This is pretty rare though.
- A couple of tests just cannot be run on Windows platforms for FIFOs do not exist or unix like permissions are not supported. A mechanism to skip them is yet to be written.
1. Setup pysh and its dependencies according to README instructions.
Note: You need to run pyshyacc.py in the pysh directory before installing pysh. That generates the pyshtables.py file (ref).
2. Clone target "crew" (see CrewRepository).
3. Clone the related patch queue from crew-w32.mq and apply it. Queue revisions changelogs contain the crew revision identifier where they successfully manually applied for the last time. They are likely to apply on children ones as well, tests evolve rather slowly.
4. Run tests as usual with "run-tests.py"
Few people actually run this procedure and you are likely to encounter setup issues. Please do not give up! Do not hesitate to send bug reports or suggestions to the author (Patrick Mézard). This tool has been used for almost two year over several Mercurial releases and has proven quite reliable.
Some relevant mailing list threads:
I was able to run the tests ("python run-tests.py -v") on crew 626cb86a6523 on Windows XP SP2 with Python 2.5.1. At first, the tests ran only up to test-archive and then hung (hg serve not wanting to be killed). I noticed that I had a stale python.exe process. After killing that manually, the tests ran through with the following summary:
Failed test-convert-git: output changed Failed test-convert-hg-source: output changed Failed test-flags: output changed Failed test-git-export: output changed Failed test-hgwebdir: output changed Failed test-imerge: output changed Failed test-merge-types: output changed and returned error code 1 Failed test-mq: output changed Failed test-patchbomb: output changed Failed test-webraw: output changed # Ran 257 tests, 26 skipped, 10 failed.
If one test hangs this is clearly a bug in the test-suite, so please report them on the usual channels. They are often caused by Mercurial not detecting it is being run in the test suite and should work in non-interactive mode. In this case, adding a -y/--non-interactive flag is usually enough to fix the test.