When saveram doesn't exist, return null, not new byte [0]. If passed a 0-length saveram file, do not pass it to the core (this crashed in interop stubs). Note that as best as I can tell, this is not fixing any problem; even without this commit it seems impossible to get BizHawk to actually generate a 0 byte saveram file for GPGX because of other checks that already existed.
* poc - have a default core menu that provides a generic settings menu item, wire up some cores to this generic menu that only have settings to begin wtih
* vectrex - remove strong type for settings, if there are no settings the convention is to use object, which allows GenericCoreConfig to be smart
* convert saturn menu to default core menu
* fixups
* convert PCFX menu to generic core menu
* convert snes9x and faust to generic core menu
* convert GBLink to generic core menus
* convert MAME to generic core menu
* convert NGP to generic core menu
* convert GGLink to generic core menu
* convert wonderswan to generic core menu
* convert virtualboy to generic core menu
Happened in a recent refactoring. NGP hides the constantfb setting (Because it is always at a constant fb), and that needs to override the core provision, not cause an exception.
Finalizers in SpeexResampler and QuickNes assume that the underlying native dll still exists when they run, as they have to in order to successfully clean up leaked memory. This is not true if those resolvers, which were stored as static fields themselves, had been finalized, which can only happen on app exit (because the static fields were readonly and always kept their value). If a SpeexResampler or QuickNes was never disposed and itself lasted all the way to app exit, then the order of these two finalizers would be unspecified and a crash could happen.
In normal circumstances, this was only observable in DS core because apparently Suuper couldn't copy paste properly and missed the dispose part, but that was already fixed in 129d454a67.
zeromus