I've started trying to fix a bug in Stanford CoreNLP and there's an issue with them accessing large arrays.
The code is 10 years old so 16GB of RAM (I'm sure) seemed a lot back them... but easy to exceed now. The limitation is that they're using a double[] to store feature vectors. I think this could be fixed by just converting all the pointers from int to long and then using a 3rd party lookup. I was thinking about using mmap but the pointers there are ints too... I could use Unsafe but that's not super fun. Do I have any other options here? I just checked and Netty ByteBuf still uses ints for pointers. So that's not an option either. One other strategy is to take a long, but internally convert it to a 2D array. This is kind of ugly though as it would require extra pointer arithmetic for each access. -- You received this message because you are subscribed to the Google Groups "mechanical-sympathy" group. To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-sympathy+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
