-
Notifications
You must be signed in to change notification settings - Fork 1.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Support for running tests on Windows #597
Conversation
* Python already has a feature for finding the temp dir. Changed test helper to take advantage of it. * Still outstanding: Several hard-coded references to /tmp appear in the tests. * Liberation-Mono is not commonly installed on Windows, and even when it is, the font has a different name. Provide a fall-back for Windows fonts. (Considered the use of a 3rd party tool to help select, but seemed overkill.)
Building/finding binaries on Windows is non-trivial. Aallow some flexibility in the path levels. (I don't want to force existing users to upgrade, but new users should be allowed the later patches.)
Doesn't do anything yet. The work is done in the subclasses that need it. Also supports context manager, to allow close to be implicitly performed without being forgotten even if an exception occurs during processes.
Demonstrate good practice in the examples.
Especially for Windows.
…led. The work should be done in close(). Deleting can be left for the garbage collector.
Again, avoid depending on __del__. Add a context manager interface. Use it lower down.
Was concerned that lambda might include a reference to reader that wasn't cleaned up by close, so changed it over to an equivalent self.reader. Probably has no effect, but feels safer.
Note: It does NOT close all the subclips, because they may be used again (by the caller). It is the caller's job to clean them up. But clips created by this instance are closed by this instance.
test_resourcereleasedemo exercises the path where close is not called and demonstrates that there is a consistent problem on Windows. Even after this fix, it remains a problem that if you don't call close, moviepg will leak locked files and subprocesses. [Because the problem remains until the process ends, this is included in a separate test file.] test_resourcerelease demonstrates that when close() is called, the problem goes away.
* Without tests changes, many of these existing tests do not pass on Windows.
Turns out Helvetica isn't sufficient. Changing to Arial (on Windows) as part of #596. This patch doesn't make the font part worse, and makes other things better, so I still recommend it. |
Helvetica wasn't recognised by ImageMagick. Changing to another arbitrary font that should be available on all Windows machines.
…sues." This reverts commit dc4a16a. I bundled too much into one commit. Reverting and reapplying as two separate commits for better history.
Removed as it was crashing on Windows, achieving nothing on Linux.
:-( I am still learning how GitHub handles PRs here. After the first three commits, I didn't expect my further commits to appear in this PR against this issue. I was going to put them in a different PR against #596. I guess GitHub expected me to use a separate branch? |
@Julian-O You are correct, separate branches result in separate pull requests. (Makes sense because in general, branch names should reflect what was/is being worked on anyway.) |
@Julian-O Not sure if you're also looking for a way to fix this, but you'd create new branches for the other PRs based on what you currently got, and could then delete individual superfluous commits (the ones belonging to this PR/branch) from them with interactive rebasing. For this one, you'd delete all commits after the last relevant one and force push the branch. |
@kerstin: Thanks for your patience. I'll need to set aside some time later to figure out how to carry out those steps. |
Closing. I'll work out how to do the branching properly in git, and submit again. |
* Exceptions do not have a .message attribute. * Help tests run on Windows - don't assume temp dir or fonts. * Python already has a feature for finding the temp dir. Changed test helper to take advantage of it. * Still outstanding: Several hard-coded references to /tmp appear in the tests. * Liberation-Mono is not commonly installed on Windows, and even when it is, the font has a different name. Provide a fall-back for Windows fonts. (Considered the use of a 3rd party tool to help select, but seemed overkill.) * Help tests run on Windows - allow some flexibility in versions. Building/finding binaries on Windows is non-trivial. Aallow some flexibility in the path levels. (I don't want to force existing users to upgrade, but new users should be allowed the later patches.) * Issue 596: Add initial support for closing clips. Doesn't do anything yet. The work is done in the subclasses that need it. Also supports context manager, to allow close to be implicitly performed without being forgotten even if an exception occurs during processes. * Issue 596: Update doctest examples to call close. Demonstrate good practice in the examples. * More exception details for easier debugging of ImageMagick issues. Especially for Windows. * Issue #596: Move away from expecting/requiring __del__ to be called. The work should be done in close(). Deleting can be left for the garbage collector. * Issue #596: Move ffmpeg_writer to using close. Again, avoid depending on __del__. Add a context manager interface. Use it lower down. * Issue #596: Update ffmpeg_audiowriter to support close/context manager. * Issue #596: Move AudioFileClip to use close(), away from __del__. Was concerned that lambda might include a reference to reader that wasn't cleaned up by close, so changed it over to an equivalent self.reader. Probably has no effect, but feels safer. * Issue #596: Support close() on CompositeVideoClip. Note: It does NOT close all the subclips, because they may be used again (by the caller). It is the caller's job to clean them up. But clips created by this instance are closed by this instance. * Issue #596: Add tests to see if this issue has been repaired. test_resourcereleasedemo exercises the path where close is not called and demonstrates that there is a consistent problem on Windows. Even after this fix, it remains a problem that if you don't call close, moviepg will leak locked files and subprocesses. [Because the problem remains until the process ends, this is included in a separate test file.] test_resourcerelease demonstrates that when close() is called, the problem goes away. * Issue #596: Update tests to use close(). * Without tests changes, many of these existing tests do not pass on Windows. * Further to PR #597: Change to Arial Helvetica wasn't recognised by ImageMagick. Changing to another arbitrary font that should be available on all Windows machines. * Issue #596 and #598: Updated test to support close(). Also changed test to meet Issue #598, but that is also being done in PR#585, so will require a merge. * Revert "More exception details for easier debugging of ImageMagick issues." This reverts commit dc4a16a. I bundled too much into one commit. Reverting and reapplying as two separate commits for better history. * Issue #599: test_6 doesn't test anything. Removed as it was crashing on Windows, achieving nothing on Linux. * Issue #596: Move comment to avoid incorporate into documents. * Issue #596: Add usages tips to documentation. * Clip class missing from reference documents. Due to failing import. * Copy-edit: Clumsy sentence in documentation. * Fix failing doctest. * Issue 596: Add initial support for closing clips. * Add key support for close() * FFMPEG_VideoWriter and FFMPEG_AudioWriter: Support close() and context managers. * Clip: support close() and context manager. Doesn't do anything itself. The work is done in the subclasses that need it. * Clip subclasses: Overrride close. * Move away from depending on clients calling__del__(). Deleting can be left to Garbage Collector. * CompositeVideoClip: Note: Don't close anything that wasn't constructed here. The client needs to be able to control the component clips. * AudioFileClip: Was concerned that lambda might include a reference to reader that wasn't cleaned up by close, so changed it over to an equivalent self.reader. Probably has no effect, but feels safer. * Update tests to use close(). * Note: While many tests pass on Linux either way, a large proportion of the existing unit tests fail on Windows without these changes. * Include changes to many doctest examples - Demonstrate good practice in the examples. * Also, migrate tests to use TEMPDIR where they were not using it. * test_duration(): also corrected a bug in the test (described in #598). This bug is also been addressed in #585, so a merge will be required. * Add two new test files: * test_resourcereleasedemo exercises the path where close is not called and demonstrates that there is a consistent problem on Windows. Even after this fix, it remains a problem that if you don't call close, moviepg will leak locked files and subprocesses. Because the problem remains until the process ends, this is included in a separate test file.] * test_resourcerelease demonstrates that when close() is called, the problem goes away. * Update documentation to include usage tips for close() Not included: * Example code has not been updated to use close(). * Merge branch 'WindowsSupport' of C:\Users\xboxl\OneDrive\Documents\MyApps\moviepy with conflicts. * Neaten up output and PEP8 compliance. Also, make runnable directly (to help debugging) * Remove references to /tmp to allow to run on Windows. * Reference to PermissionError failing on Python 2.7. * Migrate to use requests to avoid certificate problems. Old versions of urlretrieve have old certificates which means one of the video downloads was failing. Also requires changes to setup.py, to come. * Clean up of dependencies. Including adding ranges, removing unnecessary entries, adding missing entries, adding environment markers, changing versions, and updating pytest parameter handling. * Simplification of Travis file - letting te setup.py do the heavy lifting Remove conditional installations repeating the rules in setup.py Remove some installation of test needs repeating the rules in setup.py Add testing of installation options. * Add Appveyor support. * Solve Issue 629.
I have been trying to run the unit-tests on a Windows 10 machine.
Here are some minor fixes to help get past some stumbling blocks.
(Currently: 44 of 53 tests passing.)