Jerry,
What I was trying to say was that if you do NOT have the LargeIntegers plugin compiled C code (either as a separate shared library/DLL/.so OR bound into the VM), then the second of these changesets may dramatically slow down DSA execution. It was not my impresssion that any of the regularly available VMs currently have the LargeIntegers plugin internally, although that is certainly a possibility with the new plugin architecture.
Cheers, Bob
I think Bob has this right. What is more, I'd vote AGAINST having the plugin internally. There is much room for improvement in speed in the straightforward (but well-executed!) implementation and I would expect large incremental changes to be possible in the near future.
Since the "fast failure" routines are present, new largeintegers runs just as fast as old squeaks largeintegers, with the plugin giving an added boost.
At least for 2.8, until largeintegerplugin is slamming fast and unchanging, we probably ought to keep it as an in-image source plugin, with DLL's and shared libraries distributed with the image.