Frameworks/Akonadi

This is a list of things I want to do in kdepimlibs/akonadi for Frameworks 5.

Akonadi repo structure

Merge client libraries and server into one repository. Main advantage is that we will be able to share more code between server and libraries.

Proposed structure:

/autotests
/doc
/src
/src/agentbase (libKF5AkonadiAgentBase.so - AgentBase, ResourceBase and other classes related to agents. Needs a better name)
/src/agentserver (do we actually use it anywhere? Should it just die?)
/src/akonadictl (akonadictl utility)
/src/asapcat (asapcat utility)
/src/common (libKF5AkonadiCommon.so - classes shared between server, core and widgets libraries and other utilities. Does NOT install any (or most of its) public headers)
/src/control (akonadi_control utility)
/src/core (libKF5AkonadiCore.so - non-UI classes, only Jobs, models, Monitor etc.)
/src/core/jobs 
/src/core/models
/src/interfaces (DBus interfaces)
/src/qsqlite (Qt's QSQLITE driver fork - better try to upstream it)
/src/selftest (based on SelfTestDialog class)
/src/server (akonadiserver)
/src/widgets (libKF5AkonadiWidgets.so - UI classes, depends on core)
/src/widgets/views
/src/xml (akonadi/xml library)
/tests/akonaditest (akonaditest utility)
/tests/fake (fake classes to be used for testing)
/tests/searchplugin (akonadi_testsearch_plugin)
/tests/testresource (akonadi_knut_resource)

Client libraries

API/ABI changes

Status	Class	Action	Comment	Branch	Assignee
TODO	ItemSearchJob	Remove deprecated methods		frameworks
TODO	ItemSync	Make private?	Does not seem to be used anywhere	frameworks
TODO	ItemView	Remove deprecated methods		frameworks
TODO	KDSignalBlocker	Remove?		frameworks
TODO	metatypes.h	Remove		master?
TODO	MimeTypeChecker	Maybe remove	This class is able to compare mime types, handling aliases, inheritance, etc. AFAIK Volker once told me this should be in Qt 5, so check whether it's there. Might be worth upstreaming, too	frameworks
TODO	PersistentSearchAttribute	Remove deprecated methods		frameworks
TODO	PluginLoader	Remove if Qt 5 plugin system is powerful enough		frameworks
TODO	ProtocolHelper	Share with Server	See the Other Plans section below	frameworks
TODO	RecursiveItemFetchJob	Optimize	Emit itemsReceived() signal from underlaying fetch jobs	master
TODO	RenameFavoriteDialog	Make private?		frameworks
TODO	ResourceSettings	Remove?		frameworks
TODO	SearchCreateJob	Remove deprecated methods		frameworks
TODO	SelfTestDialog	Make a standalone application	SelfTestDialog is the only user of QtSQL in client libraries. By making it a stand-alone app, we can get rid of one dependency from libKF5AkonadiCore.so	master
TODO	StatisticsProxyModel	Remove		frameworks
TODO	SubscriptionJob	Split to CollectionSubscribeJob and CollectionUnsubscribeJob?	Keep the naming consistent	frameworks
TODO	TrashFilterProxyModel	Remove?	Does not seem to be used anywhere	frameworks

Features

Features can mostly be implemented already now in master, but they are something I'd totally love to see in Frameworks.

Status	Action	Comment
TO DO	Use ImapStreamParser in client jobs	Will improve performance and memory consumption on client side. Also it's easier to use :)	<{{{3}}}>
TO DO	(IF POSSIBLE) Share external payload parts helper	Concentrate external parts support into one place for easier maintanence and less code duplication	<{{{3}}}>
TO DO	Direct streaming of external payload parts	For parts bigger than 4KB, the server can instead of asking client to provide the data send back a file descriptor that client will use to write the data directly to the file. Server can either verify that sizes match, or rely on client library telling the truth. Memory--, performance++	<{{{3}}}>

Server

Status	Action	Comment
TO DO	Upstream QSQLITE3 driver	I managed to figure out what the actual changes we have in our fork are. They are not implemented optimally, a proper implementation will be more complicated, but generally should ensure better performance. Then we can try to upstream those changes and get rid of the driver for good.	<{{{3}}}>
TO DO	Remove Nepomuk support	We have Baloo now, we don't need Nepomuk anymore	<{{{3}}}>
TO DO	Remove SELECT command	See comment at CollectionSelectJob in the table above	<{{{3}}}>
TO DO	Remove backward-compatibility with older protocol versions	Some of the handlers still have several paths for supporting older client libraries. We don't even know whether they still work, so just get rid of them.	<{{{3}}}>
TO DO	Merge NotificationMessages	Merge NotificationMessageV2 and V3 to NotificationMessage	<{{{3}}}>
TO DO	Remove legacy client support from NotificationManager	Don't use the broadcast notify() signal anymore, only support notification subscribers.	<{{{3}}}>
TO DO	Refactor DB access	Try to do as much DB interaction as possible in DataStore, so that it can be easily replaced by a FakeDataStore for testing purposes.	<{{{3}}}>
TO DO	Document ASAP	Properly document all ASAP commands and responses. Ideally should be part of doxygen documentation rather than a plaintext file.	<{{{3}}}>
TO DO	Optimize CacheCleaner	Right now the CacheCleaner requests all collections every minute and checks their cache policies. Instead we should use some form of scheduling queue.	<{{{3}}}>

Other Plans

Testable Handlers

Does not really depend on Frameworks, can be done at any time. I 'want' to have it done in Frameworks though. Currently we have no way to test Handlers on the server. This would require having a fake DataStore, fake NotificationCollector, fake AkonadiConnection and somehow fake backend for Entities. However, as a result we should be able to send various commands to the handlers and observe whether correct data are stored in database and correct signals are emitted. Big problem with external payload parts, would require a fake ExternalPartManager too.

Shared protocol helper

In a perfect world, we would have Handlers in a shared library and the Handler would be able to parse the command into entities and vice versa. But since the server is using different entities than the client libraries (Akonadi::Server::Collection vs. Akonadi::Collection), this probably won't be possible. However we still should try to share code for handling ImapSets, parsing attributes, payload parts, etc. The first step toward this is to have client jobs use ImapStreamParser.