Re: [PATCH 3/3] Allow removal of current Slab

[Paul Durrant <xadimgnik@xxxxxxxxx>]

On 08/06/2023 08:10, Owen Smith wrote:
>       if (Cache->Cursor == &Slab->ListEntry)
> -        Cache->Cursor = &Cache->SlabList;
> +        Cache->Cursor = Slab->ListEntry.Blink;
>
> Looks like this blind walk back could end up with Cursor pointing to a 
> full Slab.
Right. That makes sense.

> Thinking about it, a more elegant solution could be to not remove the 
> empty Slab if its the current slab (removes the need to destroy an empty 
> slab, probably shortly followed by adding a new Slab when the nex 
> CacheGet is called). The teardown case where CacheSpill(Cache, 0), on 
> the CacheCreate fail path and CacheDestroy should NULL the Cursor, 
> effectively ignoring the Cursor == &SlabList check
> Let me try a patch that implements this method
Yes, that sounds much better.

  Paul