[sword-devel] Normalize the search string (comparing front-end apps)

[David Haslam]

Thanks, DM.

My question was not about case-sensitivity, but about Unicode normalization.
The main issue is composition vs decomposition and the canonical ordering of diacritics in each glyph.

e.g. Suppose the module contains 181 instances of the name "Efraím" which has 6 characters.
Suppose a user enters in the search box instead "E f r a i ́ m" - (NB, remove the spaces!)
That's 7 characters when normalized to NFD, the acute accent now being a separate character (U+0301 COMBINING ACUTE ACCENT).

In each front-end, will the search function find all the 181 instances (as Eloquent does) ?
Or (as with Xiphos) will it find none?

DM, what does BibleDesktop do here?

Best regards,

David

PS. ProtonMail converts automatically to NFC even though the text was keyed in as NFD, hence the above kludge with spaces.

Sent with [ProtonMail](https://protonmail.com) Secure Email.

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On 22 March 2018 1:14 PM, DM Smith <dmsmith at crosswire.org> wrote:

> It doesn’t matter that a search doesn’t use Lucene. The principle is the same. The search request has to be normalized to the same form as the searched text. For example a case insensitive search normalizes both to a single case. If it isn’t done, even on the fly, then search will fail at times. As they say, “even a blind squirrel gets a nut sometimes."
>
> Regarding Lucene there are mulitple different analyzers (that’s what does the normalization in Lucene). Each normalizes differently. Each has it’s own documentation. The analyzer that SWORD uses is suited and was developed for English texts. It is not appropriate for non-Latin texts. There is a multi-language analyzer that is much better, ICUAnalyzer, which follows UAX #29 for tokenization. For details see: https://issues.apache.org/jira/browse/LUCENE-1488 You’ll note that I participate in its development.
>
> The osis2mod proclivity for NFC is for display.
>
> DM
>
>> On Mar 22, 2018, at 8:19 AM, David Haslam <dfhdfh at protonmail.com> wrote:
>>
>> Thanks DM,
>>
>> Not all searches make use of the Lucene index !
>>
>> e.g. In Xiphos, the advanced search panel gives the user a choice of which type of search.
>> Lucene is only one of these mutually exclusive options.
>>
>> btw. Where is it documented that the creation of a Lucene search index normalizes the Unicode for the index?
>> Do we know for certain that this would occur irrespective of whether normalization was suppressed during module build?
>> i.e. With osis2mod option   -N do not convert UTF-8 or normalize UTF-8 to NFC
>>
>> Best regards,
>>
>> David
>>
>> Sent with [ProtonMail](https://protonmail.com/) Secure Email.
>>
>> ‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
>> On 22 March 2018 10:20 AM, DM Smith <dmsmith at crosswire.org> wrote:
>>
>>> The requirement is not that the search is normalized to nfc but rather that it is normalized the same as the index. This should not be a front end issue.
>>>
>>> Btw it doesn’t matter how Hebrew is stored in the module. Indexing should normalize it to a form that is internal to the engine.
>>>
>>> — DM Smith
>>> From my phone. Brief. Weird autocorrections.
>>>
>>> On Mar 22, 2018, at 5:22 AM, David Haslam <dfhdfh at protonmail.com> wrote:
>>>
>>>> Dear all,
>>>>
>>>> Not all front-ends automatically normalize the search string to Unicode NFC.
>>>> e.g.
>>>>
>>>> - Eloquent does
>>>> - Xiphos does not
>>>>
>>>> The data is incomplete for this feature in the table in our wiki page.
>>>> https://wiki.crosswire.org/Choosing_a_SWORD_program#Search_and_Dictionary
>>>>
>>>> Please would other front-end app developers supply the missing information. Thanks.
>>>>
>>>> Further thought:
>>>> For front-ends that also have an Advanced search feature, would it not be a useful enhancement to have a tick box option for Search string normalization?
>>>> Then if we do make any Biblical Hebrew modules with custom normalization, search could at least still work for the "corner cases" in Hebrew, providing the user gave the proper input in the search box.
>>>>
>>>> cf. The source text for the WLC at [tanach.us](http://tanach.us/) is not normalized to NFC, but our module is.
>>>> I'll refrain from going into a lot more detail here. There's an issue in our tracker that covers this.
>>>>
>>>> Best regards,
>>>>
>>>> David
>>>>
>>>> Sent with [ProtonMail](https://protonmail.com/) Secure Email.
>>>
>>>> _______________________________________________
>>>> sword-devel mailing list: sword-devel at crosswire.org
>>>> http://www.crosswire.org/mailman/listinfo/sword-devel
>>>> Instructions to unsubscribe/change your settings at above page
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.crosswire.org/pipermail/sword-devel/attachments/20180322/0d00a00b/attachment.html>