Sorry about being inactive - weekend completely booked. But glad to hear that you have located the issue, Niclas!
Could it be that indexing-sql tries to access type information about compoties "too soon"? If the core does the type adding in some place after extension-related code has already run... ?
On 10.4.2017 15:34, Paul Merlin wrote:
Awesome! Sorry if I contributed to that mess:/ Le 2017-04-10 13:49, Niclas Hedhman a écrit :I think I have located another "problem" in the indexing-sql implementation. And it has probably never worked. I am looking atorg.apache.polygene.test.indexing.AbstractQueryTest#script29, which triesto search for a person that has "http" field in the URL of its personalWebsite. The startup code correctly traverse the types and sets up tables for all the qnames. But during Indexing, I am getting this; WARN o.a.p.i.s.s.s.SQLCompatEntityStateWrapper - Unsupported Property type: public abstractorg.apache.polygene.api.property.Property<org.apache.polygene.test.model.URL>org.apache.polygene.test.model.Person.personalWebsite()which is the "reason" why properties within that Value type are not indexed.Ok, so what is causing this?org/apache/polygene/index/sql/support/skeletons/SQLCompatEntityStateWrapper.java:64retrieves the valueType() from propertyDescriptor and onorg/apache/polygene/index/sql/support/skeletons/SQLCompatEntityStateWrapper.java:91it checks if that is an instanceof ValueCompositeType, which it is not. Q.E.D But why is that? For instance, the propertyDescriptor for @Optional Property<Address> address(); is correct. But @Optional Property<URL> personalWebsite(); is invalid (URL is a test specific type and not java.net.URL). And the difference between Address and URL is that the former extends ValueComposite and latter doesn't. So, the reason seems to be exactly POLYGENE-137, i.e. fixed in RDF (elsewhere?) but wasn't fixed for indexing-sql. Unfortunately, no notes left behind on POLYGENE-137. And Paul's commit 7c2814ee145e91088ab6859147ef41c1d1ef8abe on 2 March 2017 mentions the POLYGENE-137 but a lot of other stuff in it at the same time.Ok, further... The ValueType that is provided, doesn't have ValueCompositein its types, as I would have expected. So why then, is ValueType notpopulated with ValueComposite, as I would have expected? Could it be that TypeLookup cuts some corners, or is it the whole model is doing something unexpected and doesn't build up all the descriptors correctly? Well, I amnot sure, and I will try to figure that out tomorrow. At least, I have located what this particular problem (script29) is all about, and there is some hope that other test cases are failing for same reason. CheersOn Tue, Apr 4, 2017 at 6:17 PM, Niclas Hedhman <nic...@hedhman.org> wrote:Thanks, I will try to dig into this. Worst case, we take indexing-sql offthe release specification. On Tue, Apr 4, 2017 at 5:26 PM, Stanislav Muhametsin < stanislav.muhamet...@zest.mail.kapsi.fi> wrote:On 4.4.2017 11:03, Niclas Hedhman wrote:Indexing-SQL scans whole application structure on startup to detect allthe visible and indexable entity composite types, and if the things changed there, it might cause problems.Oy! Do you remember How/Where is that done? Because the EntityComposite type is added to the EntityDescriptor and must be done that way (or the Entity object with instanceof) and not be checking the Java interfaces.That could be the root of many problems.It is "constructApplicationInfo" method in "extensions/indexing-sql/src/main/java/org/apache/polygene/index/sql/support/skeletons/Abs tractSQLStartup.java".However, I am looking at it (via GitHub - still no Java IDE here), and it doesn't _seem_ to use direct type check - it uses EntityDescriptors instead.You still might wanna make sure that it actually works as needed.Unfortunately, I couldn't find the code which emits "unsupported propertytype" messages (no text-content-search in Github :( ), but putting a breakpoint in there might reveal something crucial.I am very certain that I didn't get any of those when I got Indexing-SQLworking back in the days.The other change is that identity is now a type and not String, and somewhere that might cause problems. Rewrite; Yeah, I have been considering this for a while. And perhaps buildsomething by leveraging QueryDSL (http://www.querydsl.com/), or at least borrow internals for it. But I concluded that it was more work than fit into 3.0 plan. But I would be really happy to get to a situation where we have "ES backed indexer", which would help a lot going forward. The RDFimplementation is also getting "old"... Cheers Niclas On Tue, Apr 4, 2017 at 3:15 PM, Stanislav Muhametsin < stanislav.muhamet...@zest.mail.kapsi.fi> wrote: Hi!Hmm yeah, it's been quite a while since I wrote that beast. :D I woulddoa lot of things differently now, but I guess we are stuck with what wehave (until someone rewrites that).There was a discussion in October 2016, and I am not sure if the issue discussed then has been resolved, or maybe is (partially) the cause forerrors.I will copy-paste my reply I wrote in October at the end of this mail.Looking at that issue link ( https://issues.apache.org/jira/browse/POLYGENE-222 ), there might be something in detection of entity composite types - looks like "the first bad commit" was something aboutentity composite types no longer needing to extend EntityComposite interface?Indexing-SQL scans whole application structure on startup to detect allthe visible and indexable entity composite types, and if the things changed there, it might cause problems. The other commits are more unfamiliar territory for me - looking at attached test report, I think the most important information is the standard output. There is a bunch of "unsupported property type" messages - areassociations and manyassociations in Polygene just Property<..> thesedays? That definetly will break things in Indexing-SQL.It is a bit hard to follow for me since so many things have changed inPolygene now (which is also a good thing - a sign of development!). Here is the copy-paste from October.It seems that the problem was Indexing-SQL not recognizing "Identity"type as primitive (something like 'String' or 'Integer'). -------- BEGIN QUOTED MAIL -------- In "extensions/indexing-sql/src/main/java/org/apache/zest/index /sql/support/skeletons/AbstractSQLStartup.java" file, there is "initTypes" method.You might want to add mapping for Identity.class in this._primitiveTypesand jdbcTypes, and also most likely this._customizableTypes.I say "might" want to, since I have really really vague memories on howthat worked, and I realized I don't have any PC right now with Java coding environment set up.The "appendColumnDefinitionsForProperty" method in the same file usesthe type mappings mentioned above to deduce what kind of column is to be created for property, which might be the issue here. Also, you probably want to modify the "/extensions/indexing-sql/src/ main/java/org/apache/zest/index/sql/support/common/QNameInfo.java" file,so that the detection whether some Java type is primitive is updated toinclude Identity. It is done just before setting this._isFinalTypePrimitive. Let us know if this is of any help to you. -------- END QUOTED MAIL -------- On 03/04/2017 19:53, Paul Merlin wrote: Stan, Niclas,Indexing SQL has been broken for quite some time. See https://issues.apache.org/jira/browse/POLYGENE-222 That issue contains the result of bisecting the history to identify whenit broke. With the Docker based testing infra it is now very easy toreproduce. Stan, you wrote that beast :) Niclas, one commit of yours broke that beast :) Could you have a look? Thanks! /Paul-- Niclas Hedhman, Software Developer http://polygene.apache.org - New Energy for Java
