[htdig-dev] FOUND IT!!!! (Was: Bug :( )

[Lachlan A. <lh...@us...>]

Greetings all,

I think I've finally found the *source* of the recursions etc in the=20
database compression.  Once 3.2.0b5 is out, I'll remove all the hacks=20
to limit explicit recursion, and to keep the cache clean...

The problem is with the freelist of pages used when the compressed=20
page is larger than a "real" page.  It is part of the same=20
environment as the rest of the database, and so shares the cache. =20
That means that writing a page can cause access to the cache, which=20
may require writing dirty pages etc.

The solution seems to be simply to make it a "standalone" database.

Can anyone see any problems with that approach?  Do we need the=20
environment for anything?

Cheers,
Lachlan

On Wed, 18 Jun 2003 22:36, Lachlan Andrew wrote:

> I've just come across a database bug :(  It was reporting
>   WordKey::Compare: key length for a or b < info.num_length
> repeatedly when I ran a large dig without  -i.
>
> I haven't tried repeating it yet, because the dig that produced it
> takes three days!!  (It uses a rather inefficient
> external_transport) I'll try to replicate it using a more
> manageable data set.


--=20
lh...@us...
ht://Dig developer DownUnder  (http://www.htdig.org)