[TYPO3-v4] Minutes from the 1st meeting of the 4.6 Release Team

[Ernesto Baschny [cron IT]]

Olivier Dobberkau schrieb am 16.03.2011 21:41:

>> * indexed_search / search API
>>
>>>> A mindmap [8] has been created to give an overview of the whole
>> project. Dmitry still aims to improve the architecture to have a
>> pluggable search API, he personally is in favour of the SphinxQL engine.
> 
> hi guys, just a short side note on this.
> 
> <rant>
> indexed search is lame duck.
> </rant>

It's not about speeding up indexed search but making a generic search
API where different implementations (like Solr) can hook into.

> minding the fact that there is a search server that is de-facto a
> standard for enterprise search.
> http://www.cmswire.com/cms/information-management/lucene-with-solr-now-the-de-facto-cms-search-engine-006348.php

While Apache Solr it's nice for the "enterprises" and performs great and
is also not so difficult to set up as one might expect, its not feasible
for small sites hosted on shared webspaces, for example. Or even bigger
companies which cannot afford (or don't want to have) one guy to look
over / maintain yet another environment (Jetty / Tomcat). TYPO3 has
always been about "PHP" and I still think these deserve a good working
search engine.

> as you might know, we have a roadmap for the release of the apache solr
> extension for TYPO3:
> http://forge.typo3.org/projects/extension-solr/roadmap
> we would be happy to discuss the mindmap here with you.
> http://forge.typo3.org/projects/typo3v4-search/wiki/Mindmap
> when can we talk about this?

I am not at this Search project, only as an interested bystander. But in
my understanding the project should focus on:

1) A generic API (interfaces) to simply:
  a) "index content" and
  b) "search the index"

2) Manager + Factory, so that other extensions can register their search
engines to the interface: be it Solr, MySQL FULLTEXT, old-school indexed
search way, Zend_Search_Lucene, or others ...

Ok, this might sound over-simplyfied, but you probably get the message.

The core also provides one Plugin with basic searching capabilities
which uses the simple "search the index" API.

As a kick-start, I would ship with the Core the extension that provides
the Zend_Search_Lucene, because this is fast, is PHP only and probably
GPL-compatible (?).

I might be over-simplifying things, but I think we should not
concentrate so much energy in re-doing the work others have done so
nicely already (like tokenization, ranking, ...). Leave that to the
specialists (Lucene / Solr / MySQL FULLTEXT).

So the input from the Solr team is very welcome to the discussion, but
we should try get a more generic approach here for the reasons I presented.

If you think this is impossible, please rise your concerns.

Cheers,
Ernesto