searching for exact title prefix gets no matches (text search issue)

[pablomme]

Problem
=======

Searching for "package blcr-dkms 0.8.2-15ubuntu1 failed to install/upgrade: blcr kernel module failed to build" does not find bug 700036 which at the time of writing has tyle "package blcr-dkms 0.8.2-15ubuntu1 failed to install/upgrade: blcr kernel module failed to build - error: ‘struct signal_struct’ has no member named ‘count’ "

Analysis
========

Probably a fti vectorisation issue; need to determine the search terms being looked up and those present in the fti vector for the bug.

[Brian Murray (brian-murray)]

This is actually a problem with Launchpad itself and its duplicate finder feature. I've made sure that the duplicate finder is turned on for blcr - it is.

Additionally, I've written an apport bug pattern for bug 760036 so if people encounter this bug they'll be redirected to that bug number rather than going through the process of filing another bug report.

[pablomme (pablomme)]

Apport now finds the bug, thanks.

[Curtis Hovey (sinzui)]

the fti (free text index) is very clever when it handles word stems and punctuation. Too clever in some cases because exact matches are not found or are not ranked as a top match :(

[Abel Deuring (adeuring)]

Searching for the "versioned package names" like "blcr-dkms 0.8.2-15ubuntu1" should work, once the fix for ug 29713 lands. The main issue was probably the bad processing of '-' in ftq().

select to_tsvector('blcr-dkms 0.8.2-15ubuntu1');
                          to_tsvector
---------------------------------------------------------------
 '-15':5 '0.8.2':4 'blcr':2 'blcr-dkms':1 'dkms':3 'ubuntu1':6
(1 row)

select ftq('blcr-dkms 0.8.2-15ubuntu1');
                             ftq
-------------------------------------------------------------
 'blcr-dkms' & 'blcr' & 'dkms' & '0.8.2' & '-15' & 'ubuntu1'
(1 row)

select to_tsvector('blcr-dkms 0.8.2-15ubuntu1') @@ ftq('blcr-dkms 0.8.2-15ubuntu1');
 ?column?
----------
 t
(1 row)

So, the version string is cut into three parts: '0.8.2', '-15', 'ubuntu1'. It might make sense to add a special tokenizer for these version strings.