These files are only required if `PNG_ARM_NEON_IMPLEMENTATION == 1` (the default). Since we can't test it here, we just have to assume it's the case on all Arm devices. (But only Arm devices.)
When CURLOPT_READDATA is specified, not only must we set CURLOPT_READFUNCTION, but we must also set CURLOPT_SEEKFUNCTION to ensure that the data can be resent if there's a redirect.
when compiling, i get this error:
FATAL ERROR : Invalid signature for function lime@hl_mouse_event_manager_register : PP_vOiiddiidd__v required but PP_vOiddiidd__v found in hdll
This should fix it
The new version no longer checks the return code of `SDL_GetWindowOpacity`, for consistency with the other getters. In theory this should be fine, because the only documented error condition is if the window is invalid, at which point none of the getters matter. If the platform doesn't support transparency, "opacity will be reported as 1.0f without error."
As noted in pngpriv.h, the alternative (assembly code) doesn't currently
work on arm64.
There's no mention of iOS or tvOS, so maybe we could remove those?
Overriding headers has always been risky, and finally we see why.
Apparently on iOS, files may be compiled in such a way that the original
SDL_config.h file is included, not Lime's custom one. It's hard to do
anything more than we've already done to manage include order, so we
need to find another option.
One such option is to make the two files as similar as possible, such
that it usually doesn't matter which one ends up being included. Then we
only set `USING_GENERATED_CONFIG_H` when it really matters.
SDL_cocoavideo.m now enforces this, as of this commit:
ec8fa57750
That commit also removed the destructors from SDL_render_metal.m, which
implies it's required on iOS too.
And before anyone asks, SDL seems to have no plans to back down on this.
