Don't autocreate user dirs immediately when importing classes.info

[ferdnyc]

Our PPA builds for 2.6.1 are currently broken in Launchpad, because I turned on documentation building which immediately crashed. The problem is that the Sphinx config does an import of classes.info to get the OpenShot version, and as soon as it's imported classes.info attempts to create all of the preferences directories in $HOME/.openshot_qt, a no-no during package builds.

I assume this is also why the unit tests are disabled, in the launchpad configs.

All of this makes the point, though, that we SHOULDN'T BE creating a ton of directories simply because a class was imported — especially not one we import everywhere, from setup.py to doc/conf.py to freeze.py to src/tests/query_tests.py. That's an unacceptable side-effect for a class that's intended as merely a metadata dumping ground.

So, these commits move the creation of directories into a function within classes.info, setup_userdirs(), which launch.py calls immediately prior to setting up the OpenShotApp instance. All of the other scripts that import the class won't call it, so they don't trigger the creation of a bunch of unnecessary settings dirs as a side-effect.

To make this work, I had to rearrange a bunch of things in classes.settings, classes.logger, classes.project, and classes.app so that they could handle the user preferences dir being entirely absent, since the unit tests need to be able to import classes.app and create an OpenShotApp instance without ever calling classes.info.setup_userdirs().

I also removed the load of classes.logger from classes.sentry — if that's going to be loaded at the very start of the launch script, it can't trigger the initialization of the logging class. It's too early. So it just does a print() of the info instead, same as how launch.py handles its own output.

And I removed all references to the mode from the GUI code, as the unit tests haven't started the GUI for some time now. window.mode will never be "unittest" under any circumstances, anymore. (app.mode is still used by the unit tests to adjust certain startup functionality, so it's still there. Also it's now set as an instance variable, so if any of the code ever does need to check the mode, it can test if get_app().mode == "unittest".)

With these changes, it's now possible to do all of these things:

• Build, install, or package the application
• Build the user manual
• Run the unit tests

...WITHOUT triggering the creation of $HOME/.openshot_qt/ and its many, many subdirs.

Unfortunately I wasn't able to arrange things so that OpenShot itself could run entirely without a user dir as a 0-footprint application. That's something I see value in (temporary files would just be written to system default tempdirs, is the idea) since it would allow use of OpenShot in all sorts of scenarios where writing to $HOME/.openshot_qt is not feasible, permitted, or desired. But, it's a bigger undertaking that shouldn't hold up these changes.

Update

I added a few more commits:

1. classes.info now captures all of the initial defaults for user paths. (Any variable with a name that ends with _PATH, and a value that matches info.USER_PATH at the start is placed into the info._path_defaults dict.)

   • That captured dict is now used in info.setup_userdirs() (replacing the hardcoded list of variables representing paths to be created)
   • The defaults can be reapplied to reset any modified info paths, by calling info.reset_paths().
   • The default value for a path can be retrieved even if the current value has been modified, by calling info.get_default_path("NAME_OF_VAR").
   • Most (hopefully all, but I can't promise that) of the rest of the code that was responsible for resetting info paths now does so by simply calling info.reset_paths().
   • Code that had to ad-hoc things like os.path.join(info.USER_PATH, "thumbnails") to work around a modified info.THUMBNAIL_PATH now instead just calls info.get_default_path("THUMBNAIL_PATH").

2. I re-worked the autosave code a bit. The project.save() method, which used to have two parameters make_paths_relative and move_temp_files, now has only one: backup_only. Setting backup_only=True implies both of the previous parameters False, and also disables some additional logic that's undesirable when making a backup. (Like the storing of current_filepath and the resetting of has_unsaved_changes, both of which were previously being done in project.save(), then immediately un-done by the autosave code that called it. Now it's never done in the first place, because backup_only=True is passed in with the call.)

3. I renamed MainWindow.clear_all_thumbnails() to MainWindow.clear_temporary_files(), because I just couldn't abide the previous name for a function that not only cleared three different types of working file, but also deleted the (previous) backup file! A name that bad was just asking for someone to make a wrong assumption, possibly with consequences.

[JacksonRG]

LGTM

[JacksonRG]

neat way to make a very readable "and not"

[ferdnyc]

Standard disclaimer about any() and all() not supporting short-circuit evaluation applies (all of the conditions in the list will always be evaluated, unlike in a multi-clause if .. and/or ... condition), but for simple, one-time tests like this one the efficiency loss is undetectable.

[github-actions]

Merge conflicts have been detected on this PR, please resolve.

[ferdnyc]

Merging this, as it should restore our Launchpad builds to working order.

[ferdnyc]

(Just waitin' on the CI, to merge.)

[ferdnyc]

Success! And now there's an openshot-qt-doc package containing the user manual, as well.

[ferdnyc]

Oops. But I have to fix Bionic, which doesn't have python3-sphinx-copybutton available. Fixed.