Until now, the solution to ensure that even large messages are fully sent out has been to flush the connection queues after each sending of a message, which is likely an unnecessary call (with unnecessary cost) in vast majority of cases, and which may block the connection from doing other work until the large message is fully sent out. This was a rather quick, hacky workaround.
Now, after the sending the message we check whether it has been sent out fully or not. If not (outbound queues are non-empty), then we send a wake-up signal to the connection event loop. The event loop thread then fetches new sd-bus timeouts and events and will see that there are pending outbound messages to process, and will process them together with any other prospective pending events, until there is nothing to process (i.e., the outbound message has been fully dispatched).
This reorganizes the layers of abstraction in the sense how `Message` depends on `Connection` and vice versa. Now, `Message` has a link to the `Connection`. This replaces the shortcut link to the low-level `SdBus` interface that the `Message` kept. The interactions from `Message` now go through `Connection` which forwards them to `SdBus`. `Connection` is now a sole owner of the low-level `SdBus` interface. This allows for future changes around `SdBus` (e.g. a change from virtual functions back to non-virtual functions) without affecting the rest of the library. `Proxy`s and `Object`s can now send messages directly without having to go through `Connection`. The `Connection` no more depends on `Message` business-logic methods; it serves only as a factory for messages.
The flow for creating messages: `Proxy`/`Object` -> `Connection` -> `SdBus`
The flow for sending messages: (`Proxy`/`Object` ->) `Message` -> `Connection` -> `SdBus`
This also better reflects how dependencies are managed in the underlying sd-bus library.
Additionally, `getSdBusInterface()` methods are removed which was anyway planned, and improves the design by "Tell, don't ask" principle.
This refactoring is the necessary enabler for other upcoming improvements (regarding sending long messages, or creds refactoring, for example).
sd_bus_match_signal() is more convenient than more general sd_bus_add_match() for signal registration, and requires less code on our side. Plus, systemd of at least v238 is required for sdbus-c++ v2, and this version already ships with sd_bus_match_signal().
* feat: add support for direct connections
* refactor: simplify a bit, change comments, extend tests
* fix: compiler warning about unused variable
* docs: add section on direct connections to the tutorial
---------
Co-authored-by: Maksim Fedyarov <m.fedyarov@omp.ru>
Co-authored-by: Stanislav Angelovič <stanislav.angelovic@protonmail.com>
* Add methods to initiate custom session bus connection
The new function helper `createSessionBusConnectionWithAddress` allows to create connection to session bus with custom address.
Signed-off-by: Alexander Livenets <a.livenets@gmail.com>
* feat: add support for session bus connection at custom address
Co-authored-by: Stanislav Angelovic <stanislav.angelovic@siemens.com>
This internally calls sd_bus_open(), which automatically selects the
system or session bus connection based on the presence and
content of a DBUS_STARTER_BUS_TYPE environment variable and
whether the calling process has root privileges.
This option is very helpful when creating services and clients that will use the system bus in production, but connect to a session
for testing.
Additional changes:
* Removed assertions null-checking make_unique() return values.
make_unique() calls new, and new is expected to throw or abort
on failure, making the assertions unhelpful.
* Corrected a typo in the ClosesAndUnrefsBusWhenDestructed
unit test for the system bus (tested the wrong function).
* sdbus-cpp: Add API to get message credentials
Signed-off-by: Alexander Livenets <a.livenets@gmail.com>
* fix: add <sys/types.h> include for gid_t and other types
Co-authored-by: Stanislav Angelovič <angelovic.s@gmail.com>
The underlying bus was thread_local, but the design assumption that Variants built on top of that instance won't outlive the thread was incorrect. In stress tests, Variants were moved (and this is completely legal) to a different thread.
Stanislav Angelovič
Stanislav Angelovic
alivenetsandStanislav Angelovic
David Leeds
lubo-svk