Note that documentation for random_max_start_delay config setting has
been changed to indicate that the setting is in seconds, and always has
been. No functional change.
Baseline for the tests is similar to the TDBStore.
Differing from TDBStore moduletests, FileSystemStore doesn't include
tests for reserved data or corrupting the blockdevice, as it corrupts
the filesystem also.
* Optimise clearing by adding `nullptr` overload. This overload means
`Callback(NULL)` or `Callback(0)` will no longer work; users must
use `Callback(nullptr)` or `Callback()`.
* Optimise clearing by not clearing storage - increases code size of
comparison, but that is extremely rare.
* Reduce ROM used by trivial functors - share copy/destroy code.
* Config option to force trivial functors - major ROM saving by
eliminating the "operations" table.
* Config option to eliminate comparison altogether - minor ROM saving by
eliminating zero padding.
* Conform more to `std::function` API.
Use tag dispatch to better handle both NetworkInterface and NetworkStack
pointers.
The previous design was intended to avoid ambiguities when presented
with a scenario like
class MyDevice : public NetworkInterface, public NetworkStack {
};
TCPSocket(&MyDevice);
// Need NetworkStack *: use NetworkInterface::get_stack or
// cast to NetworkStack?
But the previous solution didn't actually work as intended. The overload
pair
nsapi_create_stack(NetworkStack *);
// versus
template <class IF>
nsapi_create_stack(IF *);
would only select the first form if passed an exact match -
`NetworkStack *`. If passed a derived class pointer, like `MyDevice *`,
it would select the template.
This meant that an ambiguity for MyDevice was at least avoided, but
in the wrong direction, potentially increasing code size.
But in other cases, the system just didn't work at all - you couldn't
pass a `MyStack *` pointer, unless you cast it to `NetworkStack *`.
Quite a few bits of test code do this.
Add a small bit of tag dispatch to prioritise the cast whenever the
supplied pointer is convertible to `NetworkStack *`.
The rather fiddly `nsapi_create_stack` template + overloads used during
socket formation don't inline their core, which is the identity operation
for `NetworkStack *` itself. Make code generation easier by having that
core be inline.
Moved the existing BufferedBlockDevice to features/storage unittests and switched it to gmock.
Added gmock-based unit tests to all other BlockDevice classes.
SlicingBlockDevice test left as a module test.
Major changes:
- Dependency to FileHandle removed from base classes
- AT_CellularDevice owns the default FileHandle and shares it with AT -classes
- Hang-up -detection moved as CellularContext::configure_hup(). Cannot be configured via CellularDevice any more.
Result on NRF52840_DK + BG96:
GCC:
Total Static RAM memory (data + bss): 29360(+296) bytes
Total Flash memory (text + data): 130660(-832) bytes
ARM:
Total Static RAM memory (data + bss): 261554(+8) bytes
Total Flash memory (text + data): 127573(-1193) bytes
IAR:
Total Static RAM memory (data + bss): 25479(+296) bytes
Total Flash memory (text + data): 102418(-527) bytes
RAM increase is because now ATHandler is no longer created with new -operator but is now member of AT_CellularDevice,
so image tool is able to count it. Actually total RAM consumption has decreased due to removed variables.
Disable coverage report filtering because of poor results.
Test name filtering pattern (-r) was used to filter coverage
report causing wrong results when filtered test name wasn't
matching the implementation source file name
string_to_pdp_type is only used in CellularContext classes and by having
the conversion method in CellularContext it can be used to check also
the non-standard Non-IP PDP type string.
ATHandler is part of our API so header file was moved under API folder and .cpp file was moved under device/ folder
ATHandler is used in both AT and PPP mode so it has been in slightly wrong place at the beginning.
Add a missing license header.
Remove semaphores and add +CRTDCP to support async operation.
Fix delete context and disconnect to execute just once.
Add support for NONIP PPD type.
Change CellularNetwork::clear() to virtual so it can be overridden.
Earlier CellularDevice has owned event queue used by cellular (state machine and ATHandler for example),
but the thread used to dispatch the queue has been owned by state machine.
This commit moves the event queue thread to CellularDevice so now the ownership of cellular event queue
is in one place.
- Earlier some variables were public even though used only internally
- Also refactored variables to the end of class definitions
- Removed duplicate _property_array from CellularDevice
- Changed _impl methods as protected
Rather than wait 10ms before breaking the other thread, and expecting at
least one event to have been run, wait for one event to be run, then
break.
Should avoid some spurious failures that have been seen.
Kevin Bracey
Michal Paszta
Darryl Green
Martin Kojtal
Rajkumar Kanagaraj
Arto Kinnunen
Kimmo Vaisanen
Anna Bridge
Hugues Kamba
Syed Fahimuddin Alavi
Kyle Kearney
Jarno Lamsa
Paul Szczepanek
Veijo Pesonen
Seppo Takalo
Michal Paszta
Maciej Bocianski
Qinghao Shi
David Lin
Mirela Chirica
Ari Parkkila
Roman Volosatovs
Antti Kauppila