LinkedSparseMatrix thread safe set

[fommil]

it would be great to be able to do parallel sets, and then effectively do streaming adjacency matrix construction as part of an Observable/RX pattern.

[fommil]

this is incredibly non-trivial...

[fommil]

hmm, to be able to use locks I think we'd need to have nodes for all the zeroeth rows and colums, i.e. a memory cost of N + M for an NxM matrix. I might create Scala feasibility of this in a scratch area.

[rorygraves]

You can do it with far fewer locks - you use a master lock to obtain lock instances.
Create two methods e.g.:

class LockController {
  def aquire(n : Int,m : Int) : (Lock,Lock) = synchronized {
     // we are in a guardian lock we can create new locks for N or M as needed.
  }

  def release(nLock : Lock,mLock : Lock) = synchronized {
    // we are in the guardian lock - assuming no more holder we can free locks as we feel.
  }
}

Obviously the N/M locks could be returned as a pair with aquire/release logic to ensure correctness.

This way you can reuse a very small number of locks,which approximates your row contention ratio rather than tracking NxM

[fommil]

main problem is that I need to rethink the structure to facilitate row/column locks. The current structure is not limited to a row or column :-/ writing tests for the new structure as we type...

[fommil]

giving up on this... too complicated!