Ted Zlatanov <tzz@lifelogs.com> writes:

> On Sat, 08 Nov 2014 16:39:48 +0800 Eric Abrahamsen <eric@ericabrahamsen.net> wrote: 
>
> EA> So unless I'm really missing something, my proposal is:
>
> EA> 1. Only provide one limit: max-size.
> EA> 2. Allow customization of prune-factor.
>
> EA> That seems like all the customization you'd need. Cap the size, and
> EA> provide a reasonable control of how often the registry gets pruned.
>
> OK. I am afraid I made the whole thing too complicated and it took
> another set of eyes to recognize it.  Thanks for that, and the fixes.
>
> Please go ahead and apply any changes you have in mind. I am sure no one
> is tuning soft/hard pruning. Simpler is definitely better.
>
> EA> That would require a change in the object signature, which would mean
> EA> some extra handling code for "upgrading". But once we're doing that, we
> EA> could also take the opportunity to add :prune-function and
> EA> :sort-function slots on the base registry object, which would be
> EA> nice. We could even change the default store filename from its "eioio"
> EA> extension to "eieio". :)
>
> Yeah, I remember that bug report about the extension of the database
> file, if we're upgrading anyway... It's a good chance to batch these
> changes.

Okay, here is Stab the First.

Most of what's happening should be evident from commit messages and code
comments. There are three patches:

1. The big one changes the database structure, combining :max-hard and
:max-soft into :max-size, and reworking the pruning routine to only make
a single pass. This also includes a new initialize-instance method for
upgrading from older versions, and the extraction of the database
version into a new defvar: see the comments on top of
`registry-db-version'.

2. The second patch changes the default filename to a eieio extension,
which necessitated more code than I thought -- `gnus-registry-read' is
split into two functions. I hope this isn't too crufty.

3. Number three introduces sorting into the pruning process. I
considered adding a sorting function as a slot on the registry-db
objects, but couldn't immediately think of why that was really
necessary. Anyway, there's a new option
`gnus-registry-default-sorting-function', which defaults to a function
that sorts by 'creation-time.


Tests and documentation all around.

Presumably there are bugs! I hope people will test.

Eric