Hey all,
nice discussion :-)
About implementation 1:
I like this one, but think this is what people already are doing using
their own StringBridge when needed, as an "unsupported hack".
In case the framework was offering me this option I would expect it to
be really smart: "escape" somehow the value if eventually some
non-null object is tokenized to my same marker string; I would also
expect the framework to be keyword-aware somehow on query parsing, but
this looks really messy.
A new proposal:
I got inspired by the "3VL" considerations described in Emmanuel's
link to wikipedia, and think backwards compatibility is nice:
add a "@IndexNullMarker" on the property, this will add an additional
Field to the index for null values:
@Field
@IndexNullMarker
String getFoo(){
would add either the Field "foo:value" or "fooIsNull:true" to the
Document, implementing real three-state logic. Eventually we could add
"fooIsNull:false" too for query consistency.
To resolve nameclashes (imagine a getFooIsNull() property we could use
some character prohibited in Java identifiers:
"foo-isNull:true"
or escape the name with additional prefixes as needed.
The Field and StringBridge API would remain as-is;
If you prefer not to add an additional @IndexNullMarker could be
dropped if you think adding this field is acceptable for all fields.
We could add an additional option to existing field, but I am thinking
of a case you don't want to index the value, only mark it if it is
null.
Also IMHO an additional annotation makes it more intuitive that you
are adding an additional Field.
In any case I would question the original reason for this improvement
to index null values as "default language"; if there is a default I
think you should better tell me which it is, or give it a value. So I
can't really think of a good reason to have different results for
"search-on-null" rather than "search-on-empty", but it could
potentially help in "entity state reconstruction" use cases, when the
entity is partially rebuild using projection.
regards,
Sanne
2008/4/21, Emmanuel Bernard <emmanuel(a)hibernate.org>:
Hey
The more I think about the feature, the less I like it.
Here is what I have written in Hibernate Search in Action
Hibernate Search, by default, does not store null attributes into the index.
Lucene does not have the notion of null fields, the field is simply not
there. Hibernate Search could offer the ability (and most likely will in the
future) to use a special string as a null marker to still be able to search
by "null".
But before you jump at the Hibernate Search team throat, you need to
understand why they have not offered this feature so far. Null is not a
value per se. Null means that the data is not known (or does not make
sense). Therefore, searching by null as if it was a value is somewhat odd.
The authors are well aware that this is a raging debate especially amongst
the relational model experts (see
http://en.wikipedia.org/wiki/Null_%28SQL%29).
Whenever you feel the need for searching by "null", you should ask yourself
if storing a special marker value in the database would make more sense. If
you store a special marker value in the database, a lot of the "null"
inconsistencies vanish. It also has the side effect of being queriable in
Lucene and Hibernate Search.
So before we jump on the boat for this feature, I would like to know if
people think it's still a good idea to offer this feature.
To answer your questions, the reason why I do not pass @Field but the raw
set of data is because the @Field.index is translated into it's Lucene
representation: some work is done.
Most people will write StringBridge implementation anyway where the null
handling will be taken care of transparently (by String2FieldBridgeAdaptor).
I think I like 1 or 3. Note that get should be changed as well. Three is
interesting indeed, rename it IndexingStragegy.
On Apr 21, 2008, at 10:07, Hardy Ferentschik wrote:
Hi Emmanuel,
what's you take on this? Just adding another String parameter will work, but
are we not getting too many parameters into the method? Wouldn't it be nicer
to pass the actual @Field annotation. I think this might make things also
clearer for the implementor of the interface.
I am also trying here to get a little into your head to understand your
ideas behind the code design - hope
you don't mind ;-)
--Hardy
------- Forwarded message -------
From: "Hardy Ferentschik (JIRA)" <noreply(a)atlassian.com>
To: hardy(a)ferentschik.de
Subject: [Hibernate-JIRA] Commented: (HSEARCH-115) Add a default value for
indexing null value
Date: Mon, 21 Apr 2008 14:04:33 +0200
[
http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-115?pag...
]
Hardy Ferentschik commented on HSEARCH-115:
-------------------------------------------
Ok, here are a few suggestions:
1. This is the simplest way. Basically just add a new property named
'indexNullAs' to @Field and @ClassBridge. Accordingly extend the FieldBridge
interface to set(String name, Object value, Document document, Field.Store
store, Field.Index index, Field.TermVector termVector, Float boost, String
indexNullAs).
2. Alternatively one could change the FieldBridge API to actually pass in
the Field annotation itself: set(String name, Object value, Document
document, Field fieldAnnotation, Float boost). This would reduce the amount
of parameters and might actually be more transparent for users implementing
custom bridges. Unfortunately, one would have to introduce a ClassBridge
interface as well in this case. I am not sure whether it is a good design
choice to pass annotation instances around.
3. We ccould also change the API into something like this: set(String name,
Object value, Document document, IndexProperties props), where
IndexProperties is just a wrapper class for Field.Store, Field.Index, ...
The drawback is that this just increases the number of classes.
Any comments?
Add a default value for indexing null value
-------------------------------------------
Key: HSEARCH-115
URL:
http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-115
Project: Hibernate Search
Issue Type: Improvement
Components: mapping
Reporter: Julien Brulin
Assignee: Hardy Ferentschik
Fix For: 3.1.0
Hi,
Null elements are not indexed by lucene then it's not easy to use a nullable
property in lucene query.
I have a TagTranslation entity in my model with a nullable property
language. In this case null is used as default language for tag translation.
Each translation may have many variations like synonyms.
Because I can specified a default value for null value in the @Field
annotation like this @Field(index=Index.UN_TOKENIZED,
store=Store.NO, default='null'), i can't search a cat tag with a default
translation like this : +value:cat* +lang:null
<pre></code>
@Entity()
@Table(name="indexing_tag_trans")
@org.hibernate.annotations.Cache(usage=org.hibernate.annotations.CacheConcurrencyStrategy.READ_WRITE)
@Indexed
public class TagTranslation implements java.io.Serializable {
private static final long serialVersionUID = -1065316566731456110L;
@Id
@GeneratedValue(strategy=GenerationType.IDENTITY)
@DocumentId
private Integer id;
@Field(index=Index.UN_TOKENIZED, store=Store.NO)
private String language;
@Field(index=Index.TOKENIZED, store=Store.YES)
private String value;
@OneToMany(cascade=CascadeType.ALL,
fetch=FetchType.LAZY)
@org.hibernate.annotations.Fetch(org.hibernate.annotations.FetchMode.SUBSELECT)
@JoinColumn(name="translation_id")
@IndexedEmbedded
private List<TagVariation> variations = new LinkedList<TagVariation>();
public TagTranslation() { }
...
</code>
</pre>
What do you think about that ?
Ps: sorry for english write, i am a french guy.
--
Hartmut Ferentschik
Ekholmsv.339 ,1, 127 45 Skärholmen, Sweden
Phone: +46 704 225 097 (m)
_______________________________________________
hibernate-dev mailing list
hibernate-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/hibernate-dev