When rollover occurs from a non-modifier key to a qukey, if we delay the release
event of that key until after the qukey's state is resolved, and if the hold
timeout is set to a fairly large value (on the order of 500ms), unintended
repeats would occur for a key that was actually only tapped. To prevent this, if
there's only one event in the queue (the press of the qukey), and we see a
release of a non-modifier key that's not the qukey, it's okay to allow that
release event to skip the queue and simply proceed as if it had been released
before the qukey was pressed.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Instead of unconditionally running the arduino-based build parts in verbose
mode, respect $VERBOSE. This cuts down on the amount of output we generate by
default, while still allowing us to have a verbose mode if need be.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
* never try to write anything to the host's disks
* read as little as possible from the host's disks
* keep source in ram
* cache build artifacts and intermediate content persistently
Most of these hacks are only necessary because Docker disk performance on macOS is...not performant
Signed-off-by: Jesse Vincent <jesse@keyboard.io>
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
The test has been through a major refactor, lifting out common parts, improving
comments, and naming.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
The new test case exercises a few corners of the system: we test that without
any layer changes, we start out on the right one. We also test that shifting to
another layer preserves the current caching behaviour. We test that the layers
are indeed looped through in activation order, rather than index order. And
finally, we test that explicitly turning off the single active layer will have
us fall back to layer 0.
We only test layer shifts, not locks and moves, because we only changed the
ordering, not how those behave otherwise.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
Previously, we used index-ordering for layers, meaning, we looked keys up based
on the index of active layers. This turned out to be confusing, and in many
cases, limiting, since we couldn't easily shift to a lower layer from a higher
one. As such, index-ordering required careful planning of one's layers, and a
deeper understanding of the system.
This patch switches us to activation-ordering: the layer subsystem now keeps
track of the order in which layers are activated, and uses that order to look
keys up, instead of the index of layers. This makes it easier to understand how
the system works, and allows us to shift to lower layers too.
It does require a bit more resources, since we can't just store a bitmap of
active layers, but need 32 bytes to store the order. We still keep the bitmap,
to make `Layer.isActive()` fast: looking up a bit in the bitmap is more
efficient than walking the active layer array, and this function is often used
in cases where speed matters.
As a side effect of the switch, a number of methods were deprecated, and similar
ones with more appropriate names were introduced. See the updated `UPGRADING.md`
document for more details.
Plugins that used the deprecated methods were updated to use the new ones.
Fixes#857.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
This change makes Qukeys require a certain minimum amount of time for a
key to be held before it is eligible to get its alternate (i.e. modifier)
value. This should help faster typists avoid unintended modifiers in the
output.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Jesse Vincent
Gergely Nagy
Graeme Geldenhuys
Michael Richters
Gergely Nagy
Jason Koh