Fix EZP-24553: no results for exact phrases with camelCase words

[joaoinacio]

JIRA: https://jira.ez.no/browse/EZP-24553

When performing an exact match for some phrases, no results are returned.

Problem:

After investigation it seems the cause of the problem is caused by the existence of "camelCase" words in the sentence (eg: eZ)

The wordDelimiter filter in the index analyzer performs splitOnCaseChange and concatenateWords ( which causes "eZ Publish Platform" to be indexed as "e / ez / z | publish | platform" );
However, the query analyzer does not perform these operations, so the terms end up being different and not matching.

Solution:

Changing either the index or query analyzers to the same camelCase splitting behavior.

ping @paulborgermans

[peterkeung]

If the alternative means that you modify the index analyzer to NOT split on camel case, then I vote for the alternative. There are definitely times where you want to find something like "BlackBerry" and not have it match "Black Berry". It is worth having to re-index to avoid bad matches when you were explicitly quoting the string.

[bgamrat]

I agree with Peter - it is better to reindex and avoid bad matches.

[andrerom]

I think I agree with @peterkeung and @bgamrat, but would like some input from @paulborgermans, he should be back next week afaik.

[joaoinacio]

Thanks for the feedback; indeed, changing the index instead makes perfect sense, so this PR should most likely be updated accordingly.

As noted, waiting for some additional input here from pb in case something was overlooked.
In any case everything looked good on my tests, besides the very minor BC (is it even one?)

[paulborgermans]

+1, I am ok with the pull request, it should indeed be consistent.

But also be aware that preferences on the analysis steps are up to the particular implementation. So camel case splitting may be turned of for certain customer projects, but active for others.

There is no big "truth" on what analysis should do out of the box/per default

[joaoinacio]

Updated w/ changes to index-time.
@paulborgermans: Thanks for the feedback, indeed the configuration may be a matter of particular needs and modified accordingly; this modifies the default behavior to be more "natural" with most common use cases.
@andrerom: Would there be any need to for doc. here?

[bdunogier]

+1