This commit updates the `otMessageGetBufferInfo()` and its related
types. In particular, this commit adds `otMessageQueueInfo` to
indicate number of messages, number of data buffers, and total
message length (number of bytes) for all messages in the queue.
This commit adds support for range-based `for` loop iteration over
`MessageQueue` and `PriorityQueue`. It adds `ConstIterator` and
`Iterator` in `Message` class. The non-const `for` loop iteration
(which uses `Message::Iterator`) works properly and is safe to use
even when the entry is removed from the queue during iteration. This
commit updates other core modules to use the new `for` loop iteration
model which helps simplify the code. It also updates the unit tests
to validate the behavior of the newly added iteration mechanisms.
This commit removes different `MessagePool` methods that can be used
to allocate a new message and combines them into one `Allocate()`
method which uses `Message::Settings`. The `Message::Settings` is
also updated to provide new constructor initializing it with a given
message priority only.
This commit updates `VerifyOrQuit()` and `SuccessOrQuit()` macros to
include the failed condition in the error message that is printed on
a failure (in addition to function name and line number where the
error happened). This commit also changes the second parameter
(`aMessage`) to in these macros to be optional.
This commit also updates unit tests to remove the second `aMessage`
string in cases where the failure can be inferred from the condition
itself.
This commit removes the extra `\n` at the end of error message strings
used in `VerifyOrQuit()` or `SuccessOrQuit()` macros in different unit
test modules. This help make the style (usage of macros) consistent
across all unit tests.
This commit removes the all-messages queue from `MessagePool`. This
simplifies the `Message` header `MessageInfo` and allow a message
to be inserted into one linked list.
This commit changes how the objects in OpenThread access each other.
It adds a template `Get<Type>()` method in `InstanceLocator`. This
method returns a reference to a given `Type` object belonging to the
OpenThread instance (e.g. `Get<MeshForwarder>()` returns a reference
to `MeshForwarder` object on the OpenThread instance). The
`InstanceLocator` is used as base class of all OpenThread classes so
every class can easily access any other object. This commit also
changes how the main instance is retrieved in `InstanceLocator` for
the single-instance case. The method `GetInstance()` directly uses the
raw buffer `gInstanceRaw`. This change helps make the `GetInstance()`
and in turn all `Get<Type>()` methods in-line. This commit also
removes the existing getters across all classes to use the new
`Get<Type>()` model.
This commit makes the following changes: It defines the public
`otInstance` as an empty opaque structure which is used by all public
C OpenThread APIs. It defines a new class `ot::Instance` (inheriting
from `otInstance) which is then used in core source files. The
functionality related to the instance is also moved/added into the
newly added `Instance` class (as class/static or member methods).
This commit new configuration option `enable-multiple-instances` and
its corresponding option `OPENTHREAD_ENABLE_MULTIPLE_INSTANCES`. When
enabled OpenThread supports handling of multiple instances. By default
this is disabled.
This commit also adds two optimizations for single instance case to
simplify the code and also help reduce memory/RAM usage:
(1) OpenThread objects/classes typically keep a reference to a higher
level object (e.g., many classes keep track of owning `ThreadNetif`)
to be able to access other objects/methods within OpenThread class
hierarchy. In single instance mode, the reference member variables are
removed and instead global functions are used to access the singleton
objects from one class to the other. To implement this, a group of
`<Object>Locator` classes are defined (e.g., `ThreadNetifLocator`,
etc.) which are then used as base class of other OpenThread classes.
(2) OpenThread objects which provide a callback/handler (e.g.,
`Timer`, `Tasklet`, etc.) have `void *mContext` member variable which
is used to keep track of the owner of the object. In single instance
mode the `mConext` member variables are removed since the owner is
expected to be a singleton and can be uniquely determined from the
callback function.
To implement this, two changes are made: First the handler methods are
modified to provide a reference to the object (e.g., `Timer` handler
will provide a `Timer &aTimer` as a parameter of its handler
callback). Second, a new base class `Context` is introduced which
hides the implementation providing an arbitrary context information. A
new static method `GetOwner(aContext)` is added to classes which own a
callback providing objects. This method help convert a `Context` to
the reference of the owner class object.
This commit contains the following:
- It adds methods to `Message` to assign a priority level to messages.
Currently there are four priority levels: High, Medium, Low, VeryLow
but this is easily changeable.
- It introduces a new class `PriorityQueue` which implements a FIFO queue
for storing messages based on their priority level in addition to oder
in which they are added.
If the priority level of an already queued message in a `PriorityQueue`
is changed, the associated queue will automatically move the message
within the queue based on its new priority level.
Internally the implementation uses a circular doubly linked-list with
an array of tail pointers associated with different priority levels.
- It adds new methods and an `Iterator` class in `MessagePool` to access
the priority queue where all allocated and queued messages from this
pool are stored. The `Iterator` class allows bi-directional iteration
through the list (from highest priority to lowest or reverse).
- This commit also contain detailed unit-test for all the newly added
features/methods.
Jonathan Hui