The build/ and patches/ directories are listed in .gitignore as they're
commonly used respectively as the meson build directory (this is the
directory name used in README.rst) and as a local patches store (this
usage pattern isn't official though, and may need to be reconsidered if
not used anymore). As those two directories are supposed to be at the
root of the source tree, restrict the corresponding .gitignore patterns
to avoid matching other locations.
While at it, sort the entries alphabetically, with the absolute path
entries separate from the entries that match any location within the
source tree.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
The unit test associates a fence with a framebuffer and makes sure
the fence is correctly signalled.
Once a fence is correctly signalled, it is released by the core, and the
underlying file descriptor closed.
The unit test however tries to write on the fence file descriptor every
time the designated Request is queued, an error which is now visible
as the return value of the fence signalling write() is checked.
Fix that by associating a fence with a framebuffer only once.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
The ::write() function used to signal a framebuffer fence in the unit
test is marked with the 'warn_unused_result'.
When building in debugoptimized mode not checking for the return
value causes issues at build time:
/test/fence.cpp:254:2: error: ignoring return value of function declared with 'warn_unused_result' attribute
Fix that by checking the ::write() return value and emitting an
error message in case the write fails.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reported-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
The Raspberry Pi pipeline handler now sets color spaces correctly.
In generateConfiguration() it sets them to reasonable default values
based on the stream role.
validate() now calls validateColorSpaces() to ensure that the
requested color spaces are sensible, before proceeding to check what
the hardware can deliver.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
This function forces raw streams to have the "raw" color space, and
also optionally makes all non-raw output streams to share the same
color space as some platforms may require this.
When sharing color spaces we take the shared value to be the one from
the largest of these streams. This choice is ultimately arbitrary, but
can be appropriate if smaller output streams are used for image
analysis rather than human consumption, when the precise colours may
be less important.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The ColorSpace from the StreamConfiguration is now handled
appropriately in the V4L2Subdevice.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
This adds a ColorSpace field to the V4L2SubdeviceFormat so that we can
set and request particular color spaces from V4L2.
This commit simply adds the field and fixes some occurrences of brace
initializers that would otherwise be broken. A subsequent commit will
pass and retrieve the value correctly to/from V4l2 itself.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The ColorSpace from the StreamConfiguration is now handled
appropriately in the V4L2VideoDevice.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add functions to the V4L2Device class to convert to and from
libcamera ColorSpace.
These functions are added to the base V4L2Device class so that they can
be shared both by the video device class and subdevices.
With the ColorSpace class, the color space and related other fields
are stored together, corresponding to a number of fields in the
various different V4L2 format structures. Template functions are
therefore a convenient implementation, and we must explicitly
instantiate the templates that will be needed.
Note that unset color spaces are converted to requests for the
device's "default" color space.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
This is so that applications can choose appropriate color spaces which
will then be passed down to the V4L2 devices.
The ColorSpace field is actually optional. If it is not set you will
get the camera's default color space.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
This class represents a color space by defining its color primaries,
the transfer (gamma) function it uses, the YCbCr encoding and whether
the output is full or limited range.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The CameraWorker class purpose was to handle acquire fences for incoming
capture requests directed to libcamera.
Now that fences are handled by the core library, it is not required to
handle them in the HAL and the CameraWorker and CaptureRequest classes
can be dropped.
Update the core in CameraDevice class accordingly to queue Requests
directly to the libcamera::Camera and set the release_fence to the
value of the FrameBuffer::fence() for streams of type ::Direct.
While at it make CameraRequest::StreamBuffer::fence a UniqueFD to ease
the management of the fences file descriptor values.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add a test for the Fence class by testing a Fence failure case, and
by testing a successfully signalled fence capture cycle.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Before queueing a request to the device, any synchronization fence from
the Request framebuffers has to be waited on.
Connect the Request::Private::prepared signal to the function that
queues requests to the hardware and call Request::Private::prepare().
When the waiting request queue is inspected, verify if it has completed its
preparation phase and queue it to the device.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add a prepare() function to the Private Request representation.
The prepare() function is used by the PipelineHandler class to
prepare a Request to be queued to the hardware.
The current implementation of prepare() handles the fences associated
with the Framebuffers part of a Request. The function starts an event
notifier for each of those and emits the Request::prepared signal when
all fences have been signalled or an optional timeout has expired.
The optional timeout allows to interrupt blocked waits and notify the
Request as failed so that it can be cancelled.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Since a queue of waiting Requests has been introduced, not all Requests
queued to the PipelineHandler are immediately queued to the device.
As a Camera can be stopped at any time, it is required to complete the
waiting requests after the ones queued to the device had been completed.
Introduce a pure virtual PipelineHandler::stopDevice() function to be
implemented by pipeline handlers and make the PipelineHandler::stop()
function call it before completing pending requests.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
In order to prepare to handle synchronization fences at Request
queueing time, split the PipelineHandler::queueRequest() function in
two, by creating a list of waiting requests and introducing the
doQueueRequest() function that queues requests to the device in the
order the pipeline has received them.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add an optional fence parameter to Request::addBuffer() to allow
associating a Fence with a FrameBuffer.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add to the FrameBuffer::Private class a unique pointer to a
Fence.
The Fence will be used to signal the availability of the Framebuffer for
incoming data transfer.
The Fence will be associated to a FrameBuffer at Request::addBuffer()
time, and if correctly signalled, reset by the core at Request queue
time.
If a FrameBuffer completes with errors, due to a Fence wait failure, the
Fence will still be owned by the FrameBuffer and it is application
responsibility to correctly reset it before reusing the buffer.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Introduce a Fence class which models a synchronization primitive that
allows to notify the availability of a resource.
The Fence is modeled as a wrapper of a UniqueFD instance where
read events are used to signal the Fence. The class can be later
extended to support additional signalling mechanisms.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Implement the D-Pointer design pattern in the Request class to allow
changing internal data without affecting the public ABI.
Move the internal fields that are not needed to implement the public
API to the Request::Private class already. This allows to remove
the friend class declaration for the PipelineHandler class, which can
now use the Request::Private API.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
[Move all internal fields to Request::Private and remove friend declaration]
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The Timer debug output does not report to which timer a condition refers
to. Fix that by printing the Timer address.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Limit the advertised ISP output sizes available to the sensor resolution in
PipelineHandlerRPi::generateConfiguration(). The user is free to configure a
larger resolution than this, and this will work. However, this stops strange
behavior in applications that use the V4L2 compatability layer to run, and
request the largest possible advertised resolution, which is much larger than
the sensor resolution.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Demote a couple of lines of logging to Debug level to reduce the verbosity of
the log output during startup.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The MediaDevice::populateLinks() function iterates over data links by
skipping interface links. This isn't very future-proof, it will break if
the kernel adds new types of links. Fix it by only considering data
links instead of blacklisting interface links.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Jean-Michel Hautbois <jeanmichel.hautbois@ideasonboard.com>
Reviewed-by: Jean-Michel Hautbois <jeanmichel.hautbois@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Use the new utils::abs_diff() function where appropriate to replace
manual implementations.
While at it fix a header ordering issue in
src/libcamera/pipeline/raspberrypi/raspberrypi.cpp.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
The abs_diff() function computes the absolute difference of two
elements. This may seem trivial at first, but can lead to unexpected
results when operating on unsigned operands. A common implementation
of the absolute difference of two unsigned int (used through the
libcamera code base) is
std::abs(static_cast<int>(a - b))
but doesn't return the expected result when either a or b is larger than
UINT_MAX / 2 due to overflows. The abs_diff() function offers a safe
alternative.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
The SharedFD::dup() implementation calls the C library dup() function on
the fd. When the SharedFD instance is invalid, this produces an invalid
fd, which is the correct behaviour, but logs an error message.
Fix it by returning an invalid UniqueFD directly when the SharedFD is
invalid. This also saves a system call, which is always nice to do.
Fixes: fcf98514cb ("libcamera: base: file_descriptor: Return UniqueFD from dup()")
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
We have some stream resolution which can provide slightly better
frame duration than what we cap (i.e. 1/30 fps). The problem with
this is CTS complains if the camera goes faster during the test
than minFrameDuration reported for that resolution. For instance,
1080p minFrameDuration:
- Nautilus : 33282000ns
- Soraka : 33147000ns
Both are less than capped minFrameDuration 1/30 fps (33333333.33ns).
This patch considers this situation and doesn't cap the
minFrameDuration if the hardware can provide frame durations slightly
better. The tolerance considered is 1% only from the cap.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Return the available sensor PixelFormats and sizes from generateConfiguration()
if the StreamRole is set to StreamRole::Raw. The existing code returns the
PixelFormats and sizes for all other StreamRole types.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
This introduces a way to set controls immediately for a capture
in ipu3 pipeline handler. It enables to apply a test pattern mode
per frame.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Jean-Michel Hautbois <jeanmichel.hautbois@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
This adds a function to set a camera sensor driver a test pattern
mode. CameraSensor initializes the test pattern mode by Off.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
The CameraSensor stores TestPatternModes as an int32_t. This prevents
the compiler from verifying the usage against the defined enum types.
Fix references to the TestPatternMode to store the value as the
TestPatternModeEnum type which is defined by the control generator.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
CameraStream originally creates FrameBuffers by
FrameBufferAllocator and thus buffers are allocated in V4L2 API.
This replaces the allocator in CameraStream with
PlatformFrameBufferAllocator. It allocates a buffer in a platform
dependent graphic buffer allocation API.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
The existing FrameBufferAllocator is not allowed to allocate a
new buffer while Camera is running. This introduces
PlatformFrameBufferAllocator. It allocates FrameBuffer using
cros::CameraBufferManager on ChromeOS and gralloc on non ChromeOS
platform. The allocated FrameBuffer owns the underlying buffer
but must be destroyed before PlatformFrameBufferAllocator is
destroyed.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
We cannot have a subclass of FrameBuffer because it is marked as final.
This adds a FrameBuffer constructor with FrameBuffer::Private. So we
can attach some additional resources with FrameBuffer through a
customized FrameBuffer::Private class.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
If there is no lens detected by the system, then we will not be able to
set the control, so we can skip processing of the list.
Furthermore, if the IPA has not set a V4L2_CID_FOCUS_ABSOLUTE control,
then a warning will be printed as the lensControls.get() will not
succeed.
Break out of the control parsing when there is no CameraLens
device, or if there is no absolute focus control set by the IPA.
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Add == and != comparison operators between two SharedFD instances, and
use them to replace manuel get() calls.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
For consistency with UniqueFD, rename the fd() function to get().
Renaming UniqueFD::get() to fd() would have been another option, but was
rejected to keep as close as possible to the std::shared_ptr<> and
std::unique_ptr<> APIs.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Now that we have a UniqueFD class, the name FileDescriptor is ambiguous.
Rename it to SharedFD.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
V4L2Camera::getBufferFd() returns FileDescriptor. However, the
file descriptor is still owned by V4L2Camera. It should rather
return an integer to represent V4L2Camera doesn't have the
ownership of the file descriptor.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Manages a file descriptor owned by DmaHeaps for a dma handle by
UniqueFD. Furthermore, DmaHeaps::alloc() creates a new file
descriptor and the returned file descriptor is owned by a caller.
This also clarifies it by changing the returned value to UniqueFD.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
The V4L2VideoDevice::open() function that takes an open file handle
duplicates it internally, and leaves the original handle untouched. This
is documented but not enforced through language constructs. Fix it by
passing a FileDescriptor to the function.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Manages file descriptors owned by V4L2VideoDevice by UniqueFD.
This also changes the return type of exportDmabufFd to UniqueFD
from FileDescriptor in order to represent a caller owns the
returned file file descriptor.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Manages a file descriptor owned by MediaDevice for a media device
node by UniqueFD.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
The file descriptor created for the remote side of the socket is passed
to the forked process, but never closed. Fix the leak.
The fix can be tested by running the unixsocket_ipc unit test under
valgrind with `valgrind --track-fds=yes ./test/ipc/unixsocket_ipc`.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
IPCUnixSocket::create() creates two file descriptors. One of
them is stored in IPCUnixSocket and the other is returned to a
caller. This clarifies the ownership using UniqueFD.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
