The build was failing with:
retro-gtk/meson.build:188:4:
ERROR: Got unknown keyword arguments "export_packages", "namespace"

Checking https://mesonbuild.com/Pkgconfig-module.html the arguments
"export_packages" and "namespace" are not present in pkg.generate, so
I removed them to make the build complete

This properly supports loading games and this avoids retrieving the AV
info if the game failed loading, matching what is stated by libretro.h
in the documentation of retro_load_game() and
retro_get_system_av_info().

They are virtually identical, so let's accept NULL in load_game() and
unconditionally unload any existing game, which is likely the right
thing to do anyway. This makes the code simpler.

This is needed for the tests to pass when we will corresctly support
game loading failure as expected by Libretro.

DeSmuME 2015 and Gambatte don't like this. It's not clear why - returning
NULL there is entirely valid, but it's possible we set those directories
too late.

Reverts commit b782c726

Rename it into RETRO_UNIMPLEMENTED_ENVIRONMENT so we can more easily
grep UNIMPLEMENTED to look for implemented API but unimplemented
features.

This is going to be easier to maintain.

Moving it out of the tests directory will help using it as a API
documentation.

This ensures the ABI will be correct for this function.

The strings are supposed to be constant and the booleans and bool, not
gboolean. This also fixes the ABI.

It is supposed to be constant in the first place and we don't transfer
its ownership.

This is clearer and safer.

This unsets the flag stating some variable got updated when getting
the new values and not when checking the flag, as stated in the
documentation of RETRO_ENVIRONMENT_GET_VARIABLE_UPDATE.

This supports explicit software rendering requests.

This avoids leaks if the core  happens to set the hardware render
interface multiple times.

We don't want to duplicate them.

This avoids declaring function prototypes in different C files. This
also helps knowing where some core functions are defined.

It doesn't exist in Libretro and it is equal to English so let's just
use English instead.

This matches the new name in Libretro instead of the deprecated one.

It is called core_assets_directory now.

This ensures we use strictly positive values when checking if
coordinates are part of the display, avoiding odd results and divisions
by 0.

This will help detecting issues.

This ensures we return 0 or 1 for booleans.

This sanitizes strings by returning NULL instead of empty strings. Doing
this with a macro avoids accidentally doing it wrongly as fixed by the
previous commit.

This fixes a crash.

The combination of GtkGestureClick+GtkEventControllerMotion was a naive
direct port from GTK3's event signals. However, it can't really work on
touchscreen: motion events never happen.

Instead, use a GtkGestureDrag to track the whole press-drag-release
sequence at once, and only use GtkEventControllerMotion when not dragging.

Drop the need to acklowledge the API's unstability, as we will
guarantee its stability.

Alexander Mikhaylenko authored Oct 27, 2018 and Alexander Mikhaylenko committed Jun 19, 2021

Remove pointer grabbing for now, we'll need to reimplement it via raw
X11/Wayland APIs later.

This simplifies and clarifies a bit the IPC error handling.

This makes the code safer and more concise.

It was forgotten on the previous refactoring.