Make solo half of split keyboards (more) usable. (#13523)

* Make solo half of split keyboards (more) usable.

Using only one half of a split keyboard (that's using the split_common
framework to communicate) is not a great experience, since several read
timeouts per scan cycle cause an unusably slow scan rate.

This change blocks all split communication attempts for 500 ms
(configurable) after an error occurs, causing the scan rate to become at
least _more_ usable, but might need some tweaking to work fully on most
keyboards. One read timeout still needs to occur after the 500 ms has
passed, and if that timeout isn't low enough, some scan cycles may still
be too slow.

* Fix lint complaint.

* Require 25 consecutive comm errors to see comms as disconnected.

The number of max errors can be overridden by defining
`SPLIT_MAX_CONNECTION_ERRORS`.

* Add comments to new defines, and ability to disable disconnection check.

Also increase `SPLIT_MAX_CONNECTION_ERRORS` to 40, since it's divisible
by most relevant numbers for the description.

* Make lint happy ...again

* Only update `connection_check_timer` when needed.

* Add new defines to split keyboard documentation.

* Move connection timeout logic to transport.c, add `is_transport_connected`.

* Use split_common disconnection logic in matrix.c.

Instead of doing more or less the same thing twice.

* Move disconnection logic to `transport_master`.

Is a cleaner implementation, and causes the scan rate while disconnected
to increase instead of decrease.

* Lint fixes.

* Lower default `SERIAL_USART_TIMEOUT` to 20 ms.

The read timeout must be low enough to not cause exessively long scan
cycles when using a solo split half. 10 ms was determined from testing
to work fine even with the slowest defined baudrate of 19200 (5 ms was
too low for that case), so 20 ms should be fine for most cases.

* Remove `SERIAL_USART_TIMEOUT` from ergodox_infinity/config.h

Was somewhat mistakenly included in an earlier PR.

* Fix building with `USE_I2C`.

* Reduce built firmware size.

Not really sure why this works, the idea was taken from tzarc's work on
split disconnection.

* Tweak and improve opt-out for split disconnection logic.

There are now two ways to opt out from this feature:
* Set `SPLIT_MAX_CONNECTION_ERRORS` to 0. This will completely disable
  the connection status checks (also affects the slave matrix reset logic in
  matrix.c, though).
* Set `SPLIT_CONNECTION_CHECK_TIMEOUT` to 0. This will only disable the
  communication throttling while disconnected. Will make the firmware
  smaller.

* Make split disconnection logic work with custom transports.

Includes a fallback implementation for keyboards using a custom
split_util.c but not a custom matrix.c (currently no such keyboard seems
to be merged, though).

* Remove unnecessary include of timer.h

Co-authored-by: Joel Challis <git@zvecr.com>

Co-authored-by: Joel Challis <git@zvecr.com>

1 files changed, 1 insertions, 1 deletions

diff --git a/quantum/split_common/transport.c b/quantum/split_common/transport.c
index a711ef85f0..bcc0261417 100644
--- a/quantum/split_common/transport.c
+++ b/quantum/split_common/transport.c
@@ -115,4 +115,4 @@ bool transport_execute_transaction(int8_t id, const void *initiator2target_buf,
 
 bool transport_master(matrix_row_t master_matrix[], matrix_row_t slave_matrix[]) { return transactions_master(master_matrix, slave_matrix); }
 
-void transport_slave(matrix_row_t master_matrix[], matrix_row_t slave_matrix[]) { transactions_slave(master_matrix, slave_matrix); }
\ No newline at end of file
+void transport_slave(matrix_row_t master_matrix[], matrix_row_t slave_matrix[]) { transactions_slave(master_matrix, slave_matrix); }