This now ensures that finalizeReq is indeed called from mrntt,
since exception-experiencing threads will post an exceptionInd
to mrntt, which will then call finalizeReq.
We added a new centralized OpenCL Compute manager. This can later
be extended to support CUDA, SyCL, etc. SMO can be configured at
build time to choose which API it will use for compute.
Moreover, the ComputeMgr allows us to register buffers which are
available to all cl_contexts.
This symbol is defined as a static member object inside of a
boost detail header. When boost headers are used in a project
that uses Boost in both the main binary as well as dlopen()'d
shlibs, the top_ symbol gets duplicated and the metadata gets
partitioned.
We use the Boost shlib to unify both the main binary and the
shlibs to use the same memory address for top_.
This involves marking the templated object call_stack::top_ as
"extern" and then declaring to Boost that we intend to use the
shlibs.
All Cologexes (both Cologexes and CologexSets) are now categorized
as Concepts.
Goals are now also Concepts -- they inherit from Concept as a base
class.
Using Concept as the base class for both Cologex and CologexSet
allows us to treat both cologexes and cologexsets the same way
when comparing in the abstract.
Since we have no choice but to access the sh_ptr<SenseApiLib> for
a lib before we can get its Qutex, we use this flag to ensure that
we can know whether the SenseApiLib data structure and its Qutex
are still valid when we enter -- i.e, we ensure that the SenseApiLib
object wasn't destroyed under our feet.
This function is the backbone for the DeviceReattacher daemon. It
assembles a list of all DA specs which are known to Mrntt, but which
haven't been successfully attached as yet, and attempts to attach
them.
This method wraps around attachAllUnattachedDevicesFromReq and supplies
it with a sh_ptr<> collection of all DASpecs parsed by the DAP parser
from the cmdline.
The initialization sequence now correctly initializes all DAP specs
given on the cmdline again.
This method now accepts a sh_ptr<vector<DeviceAttachmentSpec>> to
tell it specifically which specs to attempt to attach.
This enables us to implement different frontends that supply it
with collections of devices from different sources (GUI, cmdline,
previously failed-to-attach/hot-removed devices, etc).
SMO temporarily initializes none of the devices from the cmdline
during this commit as we transition to implementing the cmdline
collection frontend.
We've renamed these now to better reflect what they do.
* attachAllSenseDevicesFromSpecsReq=>attachAllUnattachedDevicesFromReq
* detachAllSenseDevicesReq=>detachAllAttachedDeviceRoles
This is also the first step in changing
attachAllUnattachedDevicesFrom to accept a sh_ptr<> to a collection
of DeviceAttachmentSpecs. This will enable us to unify the underlying
spec attachment logic and just create several front-ends for attaching
specs from multiple sources.
This performs a more complete device initialization and attachment
sequence. We'll do the corresponding teardown in the shutdown
sequence later.
We might probably do it as deviceRoleGoneAwayInd()
We've decided to add a separate notion of a DeviceRole to track attached
device roles now. We no longer use the collection of deviceSpecs to
track which roles have been attached. Rather, this list will simply
collate all known deviceAttachment specs which are expected to be
maintained in an attached state.
SMO can periodically scan through these and cross-reference this
collection with the collection of attachedDeviceRoles. Then it can
re-try to attach those which aren't currently attached at any given
moment. This will give resilience against device attachment failures
or device resets/malfunctions, at runtime.
This is logically cleaner and it begins preparing our next set
of restructuring changes. To wit: we're revamping the device
manager to distinguish between devices and their roles.
They are posted to Marionette.
* We also fixed callOriginCb invocations;
* Also made posted CBs use std::bind instead of greedily
early-invoking the CB on the servicing thread's stack.
We no longer need them because we now have
mrntt::mrntt.finalizeReq(), which does a more holistic job of
shutting down Marionette (and thus, ultimately, Salmanoff).
If even one step in Body.initializeReq was executed at all, then
whether or not it succeeded, we consider the body component to have
been initialized, at least with respect to whether finalizeReq
ought to run.
We now run body.finalizeReq even if body.init wasn't called. We'll
do a finer-grained check on each aspect of Body that needs to be
finalized now. This check was too large-grained.
This makes the initialization sequence much cleaner and conceptually
well encapsulated.
We also now dynamically allocate the Mind objects. They're allocated
dynamically by Mrntt inside of initializeReq. This means that we no
longer have to worry about jolting and cleaning up the running threads
of global mind object even when we never explicitly called
Mind.initializeReq.
Along with other conceptual improvements to our abstractions, this
patch also gets us to a real "end of program initialization" point
for the first time.
We now allocate globalMind locally inside of marionetteMain. Why?
Before now, we had an asymmetric threading situation where the
globalMind's threads were initialized at during global constructor
invocation and not on demand. This meant that we had to shut down
those threads even if we had never got to the point of calling
Mind::initializeReq.
This significantly complicated our shutdown sequence since we had
to factor in the lifetime of the std::thread objects inside of the
ComponentThreads which were inside of the globalMind object.
Now, if we hadn't called Mind::initializeReq, we don't have to
perform any Mind::finalizeReq or adjacent operations. Shutdown is
symmetrically mirrored against the operations we actually performed
during execution.
We introduced some complexity by splitting ComponentThreads into
two derivative types (MindThread and MarionetteThread) but I think
in the long term we'll be able to massage this split into a much
cleaner situation overall.
hayodea
latentprion