The plugin was both more complex and less accurate than it could have been. For
simplicity, it used a weighted average, with each cycle getting twice the weight
of the previous one. As a result, the reported average really only took into
account the last three or four cycles. On a keyboard with LEDs, some cycles
take much longer than others because of relatively rare updates, so this could
lead to misleading results, with the "average" cycle time usually being reported
as lower than it really should have been, and occasionally much higher.
This new version computes an evenly-weighted mean cycle time for each interval,
and runs more efficiently, by dividing the total elapsed time by the number of
cycles that has passed since the last report, rather than computing the time for
each individual cycle.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
We'd like to be able to set the default LED mode via Focus, so it can be
configured via Chrysalis. However, we may not want auto-save, so make that
configurable too.
To preserve the EEPROM layout, the highest bit of the previous led mode index
setting was repurposed for the auto save setting. This lets us set the default
mode to anything between 0 and 126 (or 127, if auto save is turned off).
While there, we also add an `onNameQuery` handler, to make it easier for
Chrysalis to detect if the plugin is available.
This addresses the Kaleidoscope parts of keyboardio/Chrysalis#846.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
I think it's reasonable to assume that other plugins won't be bad actors and
remove an active Turbo key from the live keys array unceremoniously, so this
check is really unnecessary.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Previously, the "sticky" state was simply ignored. Now it's handled properly,
leaving the "sticky" active Turbo key in the live keys array.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Our lookup table should have 32 entries, not 31, as Kaleidoscope-Ranges gives
DynamicMacros 32 entries.
Thanks @gedankenexperimenter for spotting this!
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
The plugin was restricted to the Model01, because it depends on a very specific
key coordinate -> geometric shape mapping. Because the Model 01 and the Model
100 share this mapping, we can safely enable the plugin for the latter, too.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
There were a number of problems with how we updated and handled our cache.
First of all, we did not support empty macros that consist only of a single
`MACRO_ACTION_END`: we returned immediately as soon as we encountered one such.
This is undesirable, we want to support such empty macros.
Seconds, we did _not_ bail out early when encountering an unknown step. We
continued reading from storage until we reached our slice's end. That's also
undesirable, because data past an unknown step is not something we can reliably
parse. We should have bailed out here.
On top of that, we did not keep our id->location map in good shape. The initial
cache update did the right thing, but if we did an update where we ended up with
less macros, our map would have dangling pointers for macro ids that no longer
exist. That's not a problem when our update clears the rest of the storage
slice, but if it doesn't, the results of trying to run an unknown macro would be
unpredictable. Even if we don't care about that, it's still very inefficient,
especially when we have large macro storage.
So, this update does a whole lot of things to improve the situation:
We now keep track of how many macros we find during a cache update, and will
refuse to play macro ids that are beyond our count. This improves efficiency,
and protects us from trying to run garbage.
We also support empty macros now, and return early from a cache update when we
encounter an unknown step type.
Fixes#1177.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
We should be treating the end of data the same way we treat a newline.
A common issue that affects a good number of plugins is that we don't deal with
trailing space well. For example, most plugins that respond with a large list of
values, where we iterate over an array or list, or something else, they usually
end up responding with a trailing space before the newline.
If we feed that same string back as an update, we can end up in a situation
where we lock up (or become very, very slow), because we want to read more data
than is available. Why? Because `Serial.parseInt()` (used by Focus under the
hood) will swallow up any leading whitespace. So if we have "255 \n" as an
argument list, we'll parse the first number, and the second `parseInt()` will
return 0, because it times out waiting for a number, consuming both the space
and the newline in the process. Thus, the next `::isEOL()` will still return
false, because `peek()` returns `-1`, signaling no data.
That can confuse the heck out of our plugins. To combat that, we should treat
end of data the same as EOL, and return false if `peek()` returns -1, too.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
When updating our map via Focus, do not read past `storage_size_`, because we do
not want to clobber storage space past our slice by accident.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
On GD32, while the build process does create .hex files, we use the .bin ones
for flashing. So copy those to the appropriate place, and do the same symlinking
as for the .elf and .hex files.
Because .bin is not compiled on all platforms, we guard it with an if that
checks for its existence.
There might be a better way to do this, but this was fast and easy.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
This matches the default EEPROM size of the underlying FlashStorage library, and
is substantially bigger than AVR-based keyboards, yet not _too_ big.
For various reasons, we're mirroring EEPROM into RAM 1:1, so we're constrained
by the size of that, too. That makes larger storage sizes undesirable at this
time. On top of this limitation, larger storage sizes also pose backup & restore
speed issues with Chrysalis, so lets settle for 16k, which is still very big,
all things considered, but not big enough to be a problem.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
The `::isSliceUninitialized()` method is required by the Base flash driver
class, but `GD32Flash` did not implement it. While the Base class does so,
`GD32Flash` is subclassing `EEPROMClass`, rather than `storage::Base`, for
historical and technical reasons. As such, we need to implement this method
ourselves.
We could probably use multiple inheritance, but I feel that would be more
trouble than its worth.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
DynamicMacros was missing a necessary `beforeReportingState()` handler that is
responsible for adding keys held by an active macro to the HID report. This
handler is identical to the one used by the Macros plugin for the same purpose.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Both Macros and DynamicMacros were only reading one byte for each `Key` object
in a tap sequence, so it would first read the flags byte of each key in the
sequence and treat it as a keycode byte, using a flags byte of `0`. As soon as
an unmodified keyboard key was encountered, this would be recognized as the end
of the sequence. This change fixes the bug by reading and using the flags byte
of each key in the sequence as intended.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Because GitHub's Ubuntu image offers multiple version fo clang-format, and
sometimes it seems to prefer a version lower than 12 (which is the minimum
required by our config file), I added an override via the new environment
variable `KALEIDOSCOPE_CODE_FORMATTER`.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
This mixes some manual work (IWYU pragmas, a better solution to the Arduino
preprocessor macros problem) with automated running of the tools. At this
point, it would be too much work to separate these into distinct commits, and
there isn't that much value to doing so.
There are still some things we could do to make things more robust, as some of
the headers need to be in a certain order, which happens to be in the same sort
order used by IWYU (`testing/*` files need to come after certain headers than
include `Arduino.h`), but it's probably not worth the clutter of adding an `#if
1` just to stop IWYU from re-ordering them.
I tried to get `#pragma push_macro("max")/pop_macro("max")` to work, but ended
up getting completely nonsensical compilation errors, so I gave up on it.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
This adds a `bin/fix-header-includes` script that uses `iwyu.py` and
`format-code.py` to manage header includes in Kaleidoscope source files. In
addition to the `src` and `plugins` trees, it is now also capable of handling
files in `testing` for the test simulator, which introduces some particular
complications due to Arduino's ill-advised `min` and `max` preprocessor macros.
The new `fix-header-includes` script can be run on a repository, and runs IWYU
and clang-format on code that differs between the current worktree and a
specified commit (default: `origin/master`), hopefully ensuring compliance with
the code style guide.
Also added: a new `check-all-includes` makefile target meant to be useful for
running as a git workflows check.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Suppose the following:
`foo.cpp` refers to the symbol `Bar`, declared in `bar.h`.
`foo.h` includes `bar.h`.
`foo.cpp` includes `foo.h`, but not `bar.h`.
`foo.h` does not refer to any symbols declared in `bar.h`.
If we process `foo.h` first, `#include "bar.h"` will be removed, causing IWYU to
fail with an error when it tries to process `foo.cpp`, but if we process them in
the other order, `foo.cpp` will get that include before it gets removed from
`foo.h`. This change sorts the files to be processed, putting all cpp files
first, then all header files, minimizing that problem.
We could do even better by saving the results from `include-what-you-use` for
every file, then going back and calling `fix_includes.py` on each of them, but I
don't think it's worth it.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
This allows `bin/iwyu.py` to get library dirs other than the ones for
Kaleidoscope when running on files in `testing`, without risk of collisions when
running on Kaleidoscope itself.
Signed-off-by: Michael Richters <gedankenexperimenter@gmail.com>
Jesse Vincent
Michael Richters
Gergely Nagy
Gergely Polonkai
Gergely Nagy
Jesse Vincent