[Exist-open] CVS Commit

[Wolfgang M. <me...@if...>]

Hi,

I have just commited my latest additions to the CVS version of eXist. I t=
ried=20
to post another message about my changes several days ago, but somehow it=
=20
didn't get through. So here's the good news:

Using the native storage backend, I have finally managed to index all my =
test=20
data (about 100MB, 5500 files) into one (!) single collection without see=
ing=20
a significant decrease in performance. Indexing speed is about 120KB/seco=
nd=20
on my machine (1600MHZ P4, JDK1.4.0).

Changes related to the native storage backend:

* Replaced simple LRU page buffers by a clock buffer implementation. Besi=
des=20
the better page prediction rate, you can now directly determine the amoun=
t of=20
memory required for each buffer from the settings in conf.xml: buffer-siz=
e *=20
page-size * 1.5 for each buffer.

* Adding a document no longer requires saving the entire collection metad=
ata.=20
Instead the new document is appended to the end of the collection, so jus=
t=20
the last page has to be in memory. Also document metadata is much smaller=
=20
now.

* Index data for the elements found in a collection is now split if it gr=
ows=20
too large (>1MB). This is a compromise between the two alternatives: addi=
ng a=20
key for each document or adding only one key for the whole collection.The=
=20
first alternative results in bad query performance for thousands of small=
=20
documents. The second results in very large index values, which are hard =
to=20
maintain.

Other changes:

* I tried to get the relational backend up and working again. Storing and=
=20
querying files should work correctly now - at least with MySQL (I still h=
ave=20
to test Postgres).

* The XPathQueryService has been extended to support nested queries, i.e.=
 the=20
query results of one XPath expression may be further processed in a secon=
d=20
query. Thus a new method XPathQueryServiceImp.query(XMLResource resource,=
=20
String xpath) has been added. Since this is an extension to the XML:DB AP=
I,=20
you have to cast XPathQueryService to XPathQueryServiceImpl.

* Updated UserManagementService: added methods listResourcePermissions() =
and=20
listCollectionPermissions().

I also tried to fix the "cannot compare null values" exception reported t=
o=20
this list. Additionally, several bug-fixes have been contributed by other=
s.

I think the changes are really worth to have a look at. It seems that the=
 next=20
release will be a good step forward. I'm currently updating the documenta=
tion=20
and I would like to release a new version of eXist during the next two we=
eks.=20
So please test the new code and send me your bug reports (and fixes ;-).

Best regards,

Wolfgang