Re: [GNUnet-developers] Re: ecrs bug

[Igor Wronsky]

On Tue, 2 Mar 2004, Christian Grothoff wrote:

> Now, there are two possible fixes.  One teaches the DBlock-insert to set the
> right bits (here in the single-bloomfilter).  Another possibility is to just
> always force insertion (and never do indexing) for files <1k. The latter is
> slightly less work to do and a better solution since none of the performance
> benefits apply for indexing files that are smaller than 1k. I've put the code
> that overrides the index option in favor of insertion for files <= 1k into
> CVS.

I agree on the 'coding cost' for the solution, but from the
viewpoint that we might (notice the conditional) like in the
future to optionally provide the clients a greater awareness of
what is currently in the datastore, this is bit of a setback. I mean,
atleast I'd like one day to see a handy interface that would
show me, in entire truthfulness, what of the local content
is still in the database, with what priority, enable me to
remove files by selecting from a list, and so on. Before this,
database.list has been some indication, but it can be trusted
a tiny bit less now.

Anyway, this is better than having the original bug around. ;)


Igor