Take note that in practice, this does not actually change the compiled code because the optimizer should be smart enough to remove the need for vtable lookups as a part of optimization (this is definitely true for MinGW). If you really prefer, I can collapse the class hierarchy into a single templated class that just has overloaded constructors - this also doesn't actually change the compiled code, but I can make the change if you wish.
[a|w|t|utf8]_from_[a|w|t|utf8_]string(xyz.c_str()) seems to be common enough to justify overloads. Also, I'm explicitly assuming that it is legal to override the NUL pointer within a C++ basic_string (e.g. - s[s.size()] = '\0'). As far as I can tell, this seems to be legal - please don't shoot if I am wrong.
* fixed target texture dimension when -intoverscan is used (this fixes the appereance of scanline and shadow mask)
* added target_scale and screen_count uniforms
* rounded corners now remain aligned with screen bounds when -intoverscan is used (single screen only)
This eliminated an unnecessary conversion step.
Also, I have no idea what this WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP) stuff is; it is hard to understand how it could possibly be correct because it ignores the 'window' parameter
This change adds id() property to input_device, which represents the
unique device id. This allows the osd layer when creating a device to
pass a friendly display name along with a unique identifier.
Currently the device id is only used to map a physical controller device
to controller id, but can be used more generally in the future. For raw
input devices, we use the full raw input name as the device id. For all
other devices, we fall back to device name as the device id. The
"uniqueness" of the device id is not currently enforced in code.
This change adds support for configuring device to conrtoller id. This
allows for stable controller ids even if USB devices are plugged /
unplugged, system is rebooted, etc.
See documentation for additional context.
