This page is here to take note of all the api break we should think about when it will be time for a Kivy 2.0
This page is not a wishlist for new features of kivy, it's here to document API design decisions that proved to be limiting or inconsistent with other parts of the framework, and will be changed. It's meant for developers to discuss such changes, not for users to propose them, please report about issues in kivy tracker instead if you feel you found such an issue.
Global
- Kivy is invasive at import: install root logger, configure few things, create window gl if needed, etc. Find a way to move to something less invasive, if possible.
- Make kivy command-line options co-exist with other projects (right now it tries to eat everything).
- Remove
KIVY_DOC
as a conditional import. We should not have to conditionally define classes.
Widgets
-
Find a way to more explicitly register widget classes if desired so we can drop the metaclass baseclass. E.g. require to decorate the class for it to be registered with factory, rather than happening automatically.
-
ModalView.background
/ModelView.background_color
inconsistency: thebackground_color
doesn't control thebackground
image.background_color
should be linked tobackground
, and introducewindow_color
for the color of the window layer.
-
TextInput:
- do_cursor_movements, do_* events need to be moved to normal events without `do_ preceding them.
-
ScreenManager:
- I (inclement) think ScreenManager.switch_to is badly named, since quite a lot of users don't read or understand the documentation and think it's the obvious way to switch screens when in reality they don't want its widget removal behaviour and should just set ScreenManager.current. This would be a breaking change though, so I'll just list it here.
-
Accordion:
- replace templates with dynamic classes, and break the api for passing templates in
-
Label:
- text_size might be more clearly named text_bounding_box or similar. Users often confuse it with font_size.
-
Templates:
- It would be nice for us to completely nuke templates and all properties related to them which seem to
cause confusion (eg.
title_args
)
- It would be nice for us to completely nuke templates and all properties related to them which seem to
cause confusion (eg.
Canvas
indexof
method might be better named justindex
, consistent with the normal list.
Kv lang
- Currently uses
hasattr(f, 'bind')
to check if it should create a binding, so other objects with a bind method (including Fbo) cause a crash. Could beisinstance(f, EventDispatcher)
instead, or maybe some other solution to not bring in EventDispatcher.
App
- App currently uses /sdcard/appname as the user_data_dir, but this is bad practice and we should really respect android's normal directory system (this is easy with pyjnius). This would probably also let us work better with android's newer multi-user functionality. Finally, I think this would probably let us remove the WRITE_EXTERNAL_STORAGE permission since this is not needed for an app to write to its own directories.
App.name
should return the application name in its intact form, currentlyclsname.lower()
is used.
Documentation
-
It'd be nice to, in the future, have a concrete deprecation/removal policy. That is, something that
stated that features deprecated in feature x will be removed entirely in version x+5 and that a certain version y will be considered a long term release such that no new features would be added to it, but bug fixes would be back-ported.More concretely, we might imagine that version 2.1.0 is a long-term release where by versions 2.1.1-2.1.x are considered bug-fix only releases. Further, features deprecated in version 2.1.0 would be marked for removal in version 2.5.0.
Gestures
- Abstract algorithms to support future changes (and remove the current kivy.gesture/kivy.multistroke redundancy)
- uix.touchanalyzer should be changed to a behavior for tracking touches
- Use JSON or other format for saving gestures, current format is not good