2008 Architectural Redesign/Design Thoughts
This is a listing, not entirely sorted, of thoughts, remarks, things to do, regarding PiTiVi's design, uses, etc...
Eventually, these ideas/thoughts/remarks ... will end up being split up for a full-size (re)design/refactor document.
They could be sorted as following:
- Coding Style : Urgent
- Core Features
- (long term) goals
- Vision (use cases ?)
- What is too slow ?
- What is not flexible enough ?
- What is not re-usable ?
- What is hard to use ?
- What is unclear ?
- What is not needed ?
- What is in the wrong location ?
- Split up code
Actual inline-review done in the 'code-review' branch here : http://gitorious.org/projects/pitivi/repos/mainline
Fundamental issues and goals
These issues/goals are not ordered, since they are all equally important.
No code shall access private values directly
- Use setter/getter and python 'property' mapper
myprop = property(_get_myprop,_set_myprop)
Core code shall NOT contain any UI-specific code (and
- This breaks the fundamental Core/UI separation
- Any code in core that has ui-specific code should be moved to pitivi.ui modules
- And the opposite for core code in the UI (ex.
All Core/UI classes can be subclassable for specific usage
- This is essential for full pitivi flexibility, for having all kinds of plugins, for having different UI, etc...
Splitup code in subdirectories/submodules
- Maybe not as much as one class per file... but that would be a good goal
- Definitely ONE base class per file though
UI is NOT compulsory ==> pitivi can be used as a LIBRARY
- There are several tools for which we need all the code from pitivi core without a UI implementation
- Ex : rendering backends, command-line/scripting tools, ....
- Disambiguate the naming somewhere
pitivi: python module/library
PiTiVi: application using the
- Still, we should have most of the UI logic in that library, but without any forced implementation (like qt/gtk/win32...)
No gnome/gtk/gobject/gst specific code in base classes
- This should go in subclasses
- Create our own event-based interface (to remove dependency on
- Johan might have ideas about this
generators/yieldwherever it can be applied
- Speedup and memory usage improvement
The only HARD DEPENDENCIES of pitivi are python core libraries and
- All other dependencies are optional and should be used from plugins
Remove the global instance code
- No code should depend on a single instance of application and project
- Needed to be able to load many projects in one instance (amongst other things)
- The majority of the pitivi module should be non-application modules
- Unit tests for pitivi core
Prepare for Python 3
- Article/presentation by Guido about this : http://www.artima.com/weblogs/viewpost.jsp?thread=227041
Not at all sorted
It seriously needs to be rethinked.
- Allow formats to create their own subclasses of core classes
- Subclassing FileSource to add format-specific information
- Subclassing FileSourceWidget to show that format-specific information in the UI
- Pluggable format support
- See Undo/Redo
- We need a native PiTiVi format
- Need to be able to save files used in the project
- Handle files/directories being moved, or loading the project on a different computer
- Simple support for playlists (ELD, PLS, ASX ?)
- Some actions might be specific to a plugin
Layer support in compositions
- This is currently implemented using hacks
- Actually use ObjectFactory for producing its content
- Remove limitation of only one brother per object
- Allow re-using indentical source/composition many times in a
- Fully synchronized (modifications to one instance are spread to all instances)
- Not synchronized (they're copies)
- Half synchronized (only some properties modifications are synchronized to all instances)
- Clarify its usage
- It contains the global information of objects you can use in the timeline
- For *any* object you can use in the timeline (source, fx, transition, composition, etc....)
- Create notion of
- They use a smaller area/time of the parent object (cutting a captured video)
- Or they could be corrected/modified versions of the original
source (i.e. they have extra effects applied to the source)
- Colour balance correction
- Audio Correction
- Media conversion (ex : Use
goomto make a video stream out of an audio track)
- The 'media-converted' ones could maybe done on the fly if you add a file of a given media type (ex:audio) into a track of anothe media type (ex:video).
- Allow easy transcoding, re-rendering or re-muxing of SubObjectFactory.
- Allow adding extra (meta)data
- Switch to a smarter and more efficient debugging system
- Use logging module ?
- do as
- Figure out how to cleanly solve the problem of plugins extending
both CORE and UI classes
- flumotion use some kind of bundles for this
SmartBin and Playground
This was originally created to cope with 0.8 issues
- Do we still need the playground ??
- SmartBin isn't flexible enough
- We might need to have more than two running at the same time
- Doesn't support being connected to more than one sink
- Need to support features like Viewing Video while recording Video AND Audio
An idea might be to have
- Can be a Timeline, live stream, File source, camera, webcam ,... (more or less like SmartBin now)
- It has a clean way of querying what it can produce
- We can connect multiple times to a stream (Audio, Video, ...) it produces
- Can be an Encoder, Hardware sinks, Raw File renderer, Network Stream renderer, ...
- Has a buffering property
Make sure the naming is coherent and comprehensible
- Get rid of the notions of 'threads' in bins (sooooo 0.8)
- Use proper naming from the editing world
- Add a glossary in the documentation
Thumbnails, captures, etc...
- Where do we store them ?
- How do we manage them ?
Hardware source and sinks
- Move plumber to core
- Have some discovery utility (i.e. several audio/video sinks)
- Have it subclassable
- Use HAL on linux
- Use ??? on win, etc....
- Have it subclassable
- Have some generic classification system (os/system agnostic)
- See SmartBin and Playground above
- needed for complex videos/operation
- Have a caching system for frame forward/backward operation
- Would act like a queue, except it would intercept seeks, do the original one, then do a second [1s before, 1s after] seek to have the data available straight away for step-by-step seeking
- For I-frame only codecs (DV, JPEG,...), we should be able to only
decode/process/encode parts that have be modified
- Would require gnonlin and core pitivi classes to support non-raw streams
- Have base classes for capture support in core
- See Hardware source and sinks
- We absolutely need support for DV and HDV
Flexible 'source provider/browser' support
Currently we only get sources from local files (even worse, it's using gnome-specific things)
Make it generic
It provides ObjectFactory objects
Allow acces to various kinds of source provider
- Local providers (filesystem, F-Spot, Gnome media (SoC project), ...)
- Network providers (Media Asset Management)
- P2P ? Using telepathy and tubes ?
Source providers might have extra information ==> ObjectFactory subclasses
Figure out what's the best way to solve this issue
- Use a source provider (let's say youtube) and grab some files
- You want to modify those files (let's say apply a colourbalance, or cut out what you need) to have SubObjectFactory
- You save your project (where/how do you save those files ? Do we leave them as a youtube link ?)
- You reload your project... but you don't need the youtube browser anymore (you've already got your sources)
- PROBLEM Disambiguate SourceProvider and SourceBrowser. I'd say the difference is that stuff from SourceProvider(s) have to be moved in the SourceBrowser before being used in the timeline.
See also the ideas in ObjectFactory above
Marker points / KeyFrames
Here keyframes are not I-frames but 'key positions'
- Add support in:
- ObjectFactory (interesting points, user-introduced data, could be stored with/along the file...)
- TimelineObject (instance specific property modification)
- Support multiple versions of one source
- Ex :
- low resolution versus HD/4K
- Distributed/P2P files (first download lowest resolution)
- Use thumbnail when file not available yet.
- Ex :
- if it's a network URI, option to save it locally
- detect if remote source has changed
- Support for separate index (save/load support)
- Add basic tests to make sure the sources will behave properly with
- ex : testing seeking support
- If not, propose transcoding, or plugin to download, etc...
Multiple Video support
There's two different use-cases in fact
- Steroscopic or multi-angle support (multiple video tracks)
- Editing together multiple videos from the same scene (one video
- You want a way to view all the input videos at the same time, synchronized, so you know which ones to select at any given point
- This of it as a non-live mixing-desk
- Have a
gst_launchplugin where users can create their own freeform effect pipeline
- Allow concatenation/coupling of existing effects into a bigger one
- Bonus for adding some scripting in it !
- Have Audio+Video effects
- They're in fact two/more separate effects that have synchronization
- Ex : photo flash effect : you want to have the 'flash' sound in sync with the 'whiteout'/still frame on the video side
Allow easy creation of (parts of) the Timeline
- Should have a clean content/Script separation
- Ex : Interview script : have the interview person's name separated from the actual video on which to overlay his name
- Allow subclassing
- Ex : You can create different ways of displaying the interviewee's name, different ways to blend in/out, ...
- Have a repository for those scripts/scenarios
- People can easily share/use/reuse existing scenarios
How do we handle this properly ?
- Text track ?
- Use case : karaoke/subtitle overlay
Missing feature support
Allow cleanly support Projects using unavailable subclasses/plugins/effects.
- Might not be available
- Might be proprietary/custom
- Older version of PiTiVi
- Provide a way to upgrade the feature/plugin
Central/Distributed repository of PiTiVi plugins
Missing (GStreamer) plugin support
Distributions should support this !
Easy transcoding of sources
- Allow transcoding sources only
- Useful also for making sources more editing-friendly
- Or to have lower resolution versions to work with
Optimize the following use cases
- By end (modifying media_duration + duration)
- By beginning (modifying media_start+start+media_duration+duration)
- Moving many objects at once
One idea for this would be to have a 'block-rebuild' property on GnlComposition that would mark down that is has (or not) to rebuild the internal stack, but postpone it until the property has been set back to False.
- interlaced support
- caps (differentiate raw frame/fields)
- Buffer Flags (TFF, Repeat)
- ... and obviously support in virtually all relevent plugins
- Perfect Profesionnal Colourspace support
- Various Subsampling + Chroma placement (right now we don't make a difference between 4:2:0 jpeg/mpeg2/dv-ntsc ... whereas they have different chroma placement)
- Various Clamping matrices (HDYC for example)
- FAST and bit-accurate converters
- This might require making a generic colourspace converter bin which searches all required/available colourspace converters
- We need more screenshots
- Propose nightly builds
These are some (long term) ideas I have
- GoogleMaps/OpenStreetMap plugin
- Use maps to show a trip
- Zoom-in/zoom-out/move from location to location between the various steps
- Could use photo geotags to automatically known where to go
- Photo Flash
- Could be a nice way to do transitions when doing slideshows
- Could have a 'flash' sound on the audio track, sync-ed with it
- Film-reel at slow-speed
- Start from a still frame/picture
- Gradually speed up the movie
- You see the 'film-reel border' moving (i.e. at some points the inter-frame black borders will be visible)
- Have a flickering soundtrack
The results of the search are