GitLab

This ensures we read only readable device files starting with event and
hence handled by evdev, and not other protocols. This avoids writing
garbage to devices that use another protocol, which could lead to messed
up input devices.

This uses has_key() and has_abs() instead of libevdev_has_event_code()
directly in *_new(), making it simpler to read.

This uses the file change event type to check for file deletion, rather
than using g_file_query_exists(). This prevents some race conditions, as
explained by the documentation of g_file_query_exists().

As g_udev_device_get_property will already return NULL if there's
no such property for the device, and as such it won't match the
g_strcmp0.

This shorten all this part a lot, and lets the thoroughly tested printf
do the job.

Fixes #23

This way it makes sense straight away that you're getting the sdl_string
for the guid.

NXIO has a specific use case, that is, no such device or address, and
isn't meant for general-purpose use. Since there isn't any
G_FILE_ERROR_NOT_A_GAME_CONTROLLER, use G_FILE_ERROR_FAILED which is
made to be used when no unix error code exists for it.

Those were used when the mapping string is wrong (i.e. user mapping
string), so it's not a programmer error (as that's what g_critical() is
made for).

It doesn't make sense to pass G_FILE_ERROR_FAILED, as this one is only
meant for when there isn't any existing error code for this file error,
but here it isn't the case since we have the errno set on error.

It wasn't used anywhere in the code.

So you don't need to look at the code to understand.

It would have made sense if there was extra things done on @self but
here it doesn't make much sense. So shorten it to help scanning the code
and quickly understanding it.

These aren't programming errors from our side nor from the library
user's side, but from some bad data.

As g_return_if_fail is only for public API checking. If there need to be
some arguments checking in static functions, it is a programmer error
and so must assert.

This commit also removes type checking for self (i.e. the type defined
in this file), since those check should already be put for public API
functions.

This tests on Fedora to allow testing with it works with gudev.

I accidentally didn't test with udev when refactoring a MR and messed up
that function.

Add the namespace_object_ prefix to the finalize function to match the
code style.

Add the namespace_object_ prefix to the finalize function to match the
code style.

This makes our database more consistent, and will help us when we will
normalize the database, as normalized mappings don't have the platform
key.

This prepares the normalization of the file.

That is, of the form <object_type_name>_<signal_name>_<cb>.

That is, of the form <object_type_name>_<signal_name>_<cb>.

That's what we do almost everywhere else and in other libs I maintain,
let's standardize on that.

This makes the functions trying to free some data simpler and hopefully
more optimized.

This simplifies the code and make it cleaner.

A mapping string is always expected to start with a 32 hex-digits field
and a text field, otherwise it's invalid, so let's check that.

Thanks to changes introduced by the previous commit, the mapping object
can always expect to be fed a complete mapping string, with he first
field being the GUID and the second one being the gamepad's name.

These two first fields must hence be skipped.

This will allow the mapping object to always expect a full mapping
string, with the GUID and the gamepad name.

It speeds up a lot the initial flatpak build. This commit also disables
the tests and documentation of evdev to speed up more the build.

This avoids having to manually build the summary, with all what it
entails (e.g. aligning label, having to use .format() extensively).
This commit also bumps meson minimum version requirement accordingly.

As it uses '/ with string arguments' and 'install arg in configure_file'