[Omake] SVN Commit: OMake Build System [0.9.8.x] (Rev. 10621)

[Aleksey Nogin]

On 03.05.2007 16:10, Jason J. Hickey wrote:

> The problem with the new implementation is that the normal operation, the
> coarse comparison, now gets a lot of collisions *by definition*.

Huh? Didn't you just solve this by making sure that the coarse hash is 
available? If the goal is to make the comparisons faster, we do not need 
to cons-hash, just to include the hash number in the data structure.

> There is a solution, which is a vertical split, where each fine
> node has the hash-consed coarse node as a component.  Comparisons
> should be based on the coarse node, but the full hash-consing would
> be case-sensitive.

What is the point of cons-hashing the coarse structure and then 
cons-hashing it _again_ inside the fine structure?

What we may want to do instead, may be, is to change the FileCase.t to 
be StringHash.t instead of string. This way we'd still get the maximal 
allowed sharing between the different case-variations of the same filename.

Aleksey