• Evaluate whether boost::intrusive::list<> is the proper class to use in the first place. Is this class causing massive template code bloat?
• Consider that the boost::intrusive::list_member_hook<> cannot be made volatile. This disqualifies its use within the timer observer/observable class structure.

If it is to be used within the BLE landscape:

• Use the boost::intrusive::constant_time_size<false> flag in the boost::intrusive::list<> template arguments. This will allow for boost::intrusive::list_member_hook<>::unlink() usage.
• Remove boost::intrusive::list::remove() from usage. This requires O(n) to perform. Use boost::intrusive::list_member_hook<>::unlink() instead for immediate results.
• Remove operator == from classes which include boost::intrusive::list_member_hook<> in their class body.

The peer manager component from Nordic needs to be understood - perhaps modified.
Bonding support for the BLE peripheral class needs to be implemented

GATT MTU size is specified several places:
nrf/ble/nordic_ble_stack_observer.cc
nrf/ble/nordic_ble_stack.h
nrf/ble/att.h: static constexpr length_t const mtu_length_maximum = 251u

Get to the bottom of the actual GATT MTU maximum and provide one source of truth.

Either hack/fix boost::intrusive::list to allow for a volatile hook or write a replacement.
Within the timer structure usage, this is not acceptable.
hooks are routinely linked/unlinked across interrupt boundaries.

Consider moving peripherals/ dir under nordic/ directory.

• ble_gap_connection::disconnect(), Line 63 It would be expected that scanning would need to be restarted on Central disconnect. However, calling start() results in Nordic error 8: invalid state.

In initialization and based on temperature changes, the Nordic SAADC calibration task needs to be run.

This only works on OSX.

• The cross compiler options for intrinsics is not working. Example: __REV() is not defined.
• The -isystem include rule does not work properly for gcc. It requires -I

A bunch of things:

• Use consistency in completion event types and use size_t where possible.
• In spim completion callback, provide address and length params - similar to the others.
• Check buffer lengths on std::equal() comparisions.

This is would be nice to clean up.
Create an public functional interface which hides the internal container.

The GATT client observer is a work in progress.
Complete the implementation of the GATT client observer implemented using the Nordic softdevice S132.

1. Consider using CMSIS peripheral interfaces when possible.
2. Move generic interface header files into a separate directory.
3. Timer & RTC interfaces contain Nordic specific instance information.

Example: The classes timer_observer_generic and timer_observable_generic are nearly generic in that they contain no external dependencies. Nope, false: They have the number 6 scattered about for the cc_index_limit which is Nordic specific.

This needs to be removed so that the interface is truly generic.

The following items need implemented:

• Key distribution
• MITM protection
• LESC authentication, key exchange
• Legacy Pairing
• GATT read/write authentication

On initialization and based on temperature change the SAADC calibration TASK needs to be run.

See this link:
https://devzone.nordicsemi.com/f/nordic-q-a/15908/nrf52-twi-excess-current-consumption/60684#60684

Errata 89 tells that this is an anomaly inside the chip. And the work around is given.
Just add below code when you want to disable TWI

For TWI0 for number 89 anomaly on nRF52

*(uint32_t *)(0x40003000 + 0xFFC ) = 0;
*(uint32_t *)(0x40003000 + 0xFFC ) = 1;

For TWI1 for number 89 anomaly on nRF52

*(uint32_t *)(0x40004000 + 0xFFC ) = 0;
*(uint32_t *)(0x40004000 + 0xFFC ) = 1;

The BLE observers for GAP, GATTC, and GATTS all must be wrapped in a Nordic wrapper in order to register. This is not necessary.
Move the intrusive list hook inside the GAP, GATTC, GATTS event observer interface.
The wrapping in Nordic will no longer be necessary.

This will allow the client to stream multiple packets without gaps in the SPI clock.
This will provide a method for streaming large packets (>255 bytes) of data without the use of EasyDMA array lists.

Most (all?) of the Nordic peripherals' deinit() methods are poorly written and likely do not work.

• Evaluate and determine the proper methods for deinit(); should these be blocking calls unit clean up is complete or async with notifications.
• Write tests against these methods.

After simplifying the segger RTT module, the remaining thing to do is provide the up/down buffers as allocations provided by the client and remove the idiodic #define methods currently employed.

Move test projects

• app_timer_test/
• rtc_test/
• spi_test/
• timer_test/

Into top level directory system_tests

So that functionality is only implemented in once place, the service container, service and characteristic descriptor_list find() methods, which are non-const, should not duplicate the implementations of their const partners; but call the const version.

This has led to bugs which I do not yet understand.
Understand them, fix them and resolve this.

When transitioning into low power state, peripherals should follow a pattern.
There is no pattern - create one.

These classes are hardware independent and would benefit from unit tests.

uint32_t volatile* rtc::cc_get_event()
uint32_t volatile* timer::cc_get_event() specific to Nordic

If possible, create a more generic interface if other silicon vendors provide this type of functionality.

The temperature sensor will provide much more accurate readings if the factory calibration data is used.

See https://devzone.nordicsemi.com/f/nordic-q-a/1948/strange-error-code-13313-0x3401-returned-by-sd_ble_gatts_hvx/8365#8365

As the error suggests, it means that the softdevice doesn't yet have a set of system attributes. :-)

I'd recommend you to read this document to better understand the concept of bonding with regard to our softdevice, since it also applies to the system attributes.

As you can see, bonding data is split between encryption keys and the system attributes. System attributes includes for example values for CCCDs, and the softdevice therefore needs to get those before sending any notification. Hence, when you try to send a notification, if the peer device have not yet written to the CCCDs, the softdevice will ask the application whether it has any previous bonding data to be used, by giving a BLE_GATTS_EVT_SYS_ATTR_MISSING. If the device does use bonding, the bond manager will normally handle this event, and pass back whatever data has previously been written.

However, the system attributes are requested with the same event no matter if the device use bonding or not (the softdevice doesn't really know that). If a device does not use bonding, but in this case, you can just reply like this:

Class is in: gap_connection_state.cc.
Also include class ble::gap::connection::negotiation_state::completion_notify

The tlv_encode functions and the ble::gap::advertising_data_t class need unit test coverage.

The implementation for discovering 128-bit UUIDs using the Nordic softdevice will work find
for single connection GATT clients. However, for multiple GATTC clients the stored struct uuid128_read_pending_t instance will conflict.

The struct uuid128_read_pending_t instance needs to be included somewhere in the ble::gattc::connectable hierarchy so that there is one instance per connection.

See ble/nordic_ble_gattc_event_observable.cc gattc_uuid128_acquire() comments.

This is probably related to the Nordic errata:

     * Errata nRF52832 Rev2, v1.0 3.44 [173]
     * GPIO: Writes to LATCH register take several CPU cycles to take effect
     * Conditions:   Reading the LATCH register right after writing to it.
     * Consequences: Old value of the LATCH register is read.
     * Workaround:   Have at least 3 CPU cycles of delay between the write
     *               and the subsequent read to the LATCH register.
     *               This can be achieved by having 3 dummy reads to the
     *               LATCH register.

The LATCH set of bits contains the bit from the previous interrupt.

The ARM GCC compiler v 7.1 does not support charconv.

Until the switch is made to 8.x we are faking out charconv in nrf/utilities/charconv.h.
Convert to GCC ARM 8.x when appropriate and remove the charconv.h fakery.