* Initial commit for the CLI settings compiler
Not very useful for now, only generates settings.c in the same
way the settings were manually written in cli.c
* Move all settings to a YAML file
This will eventually let us compile and pack the settings saving
a lot of memory. For now, the code compiles but it doesn't work
since it uses a byte to index into the word array which has more
than 256 entries.
* Use varint encoding for cli name word indexing
This makes the CLI work again.
* Make clivalue_name_* funcs return bool
Makes more sense than returning uint8_t, even when the compiler will
probably generate exactly the same assembly in both cases.
* Fix invalid field name
Missing a closing ]
* Initial attempt at generating the settings files at build time
Optimize the generator to call into the compiler only once, so
we can afford to call it for each build and, eventually, generate
build-optimized settings.
* Fix build error due to generated files
Due to make's expansion rules, the generated implementation file
wasn't correctly compiled if the build was started when the
generated files didn't exist.
Althogh there's probably a better solution, this should work for
now.
* Generate a per-build settings_generated.{h,c}
This allows us to save a bit more space, since this way the words
array doesn't include words which are not used by the build.
* Remove pgn_t field from cliValueConfig_t
Use a couple of arrays to find the pgn_t for a setting from its
offset in the table. This saves another 384 bytes on NAZE.
* Use only a byte for the field offset in clivalue_t when possible
While compiling the settings, determine if any offset requires a
number bigger than 255. If that's not the case, use a uint8_t
rather than an uint16_t for storing the field offset.
* Add missing header to PG_MODE_ACTIVATION_OPERATOR_CONFIG group
* Fix unbalanced #endif
Introduced when deleting the hardcoded settings from cli.c
* Don't ignore the return value from g.CanUseByteOffsetoff()
CLIVALUE_USE_BYTE_OFFSETOF was always defined regardless of the
maximum offsetof() value found in the settings.
* clivalue_name_*() functions now take a buffer
Requires only CLIVALUE_MAX_NAME_LENGTH bytes in the stack rather
than 2*CLIVALUE_MAX_NAME_LENGTH, since those functions were called
from functions which already had a buffer for the name allocated
but had to allocate their own.
* Remove unneeded clivalue_get_name() call
clivalue_name_exact_match() will already fill the buffer with
the value name.
* Fix off-by-one error in the settings generator
The generated C code wasn't allocating enough space for the '\0'
terminator for setting names
* Fix off-by-one error in the name decoder
CLIVALUE_ENCODED_NAME_MAX_BYTES represents the maximum number of
bytes in an encoded name, not the maximum word index.
* Add missing headers to PG_STATS_CONFIG group
* Make sure the settings are always up to date
* Initial attempt at encoding constants used for min/max settings
Pretty naive approach for now. Saves ~400 bytes on F1 targets.
* Move tool for generating settings to tools/
Also, rename it from settings_gen to just settings.
Delete the .gitignore in src/main/fc and just add all ignored
files in the root .gitignore, since that speeds up git.
* Only print setting stats when the env var V=1
This way we get quiet output unless the Makefile has been invoked
with verbose output.
* Make setting generation rules compatible with gmake 4
Rules were working fine on gmake 3, but failing with gmake 4. These
new rules should work with both of them.
* Fix constant value detection with GCC 7.1
GCC 6.3 emits errors with <42type-suffix> while GCC 7.1
emits the errors with only <42>
* Format uint8_t arrays a bit better
Don't add a comma after the last element
* Sort words and values determiniscally
This will help while checking the upcoming Ruby implementation
of the generator against the previous one using Go.
* Add missing headers for some groups in settings.yaml
* Replace the Go settings generator with a Ruby implementation
This makes it easier to install the required dependencies to
build INAV, since Ruby is installed by default on macOS and
installing it in Linux should be easier than installing Go
and a 3rd party package (for YAML parsing).
* Don't hardcode the value type for each parameter group
Instead, add a value_type field to each group with a default
value of MASTER_VALUE
* Simplify code for adding custom methods to StringIO
* Only resolve types for enabled fields
This fixes issues with some types which are only defined
if the feature for them is enabled (e.g. STATS or NAV).
* Implement print_stats() in the Ruby settings generator
* Rename constant in generated settings
CLIVALUE_ENCODED_NAME_USES_DIRECT_INDEXING =>
CLIVALUE_ENCODED_NAME_USES_BYTE_INDEXING
* Remove old settings generator binary from .gitignore
* Enable DEBUG while generating settings
Travis build is failing, this should help determine why
* Add $TOOLCHAINPATH to $PATH on Travis builds
* Disable DEBUG in settings.rb
Travis build is now failing because the log is too big
* Fix warning when running settings.rb on RB >= 2.4
* Don't print message when generating settings with V=0
* Use a relative path for the temporary dir
Absolute paths cause issues calling out to g++ on Windows
* Add INAV license header to settings.rb
* Add missing header to settings.c
Required since last rebase, it was compiling fine previously
* Remove unneeded extern variable decl from settings.c
Not needed anymore since we're now including settings_generated.c
directly in settings.c to simplify the Makefile.
* Use obj/tmp rather than just tmp for temporary files
* Update devdocs to mention Ruby installation
* Update Dockerfile and Vagrantfile to install Ruby
Required by the settings generator
Fixes#1997
Niccolò Maggioni
Andy
Alberto García Hierro
Aleksandr Kuzmitsky