We'd deliberately changed the tracker from searching issue texts by 
default to searching issue titles.

We did this because we were returning so many irrelevant results that we 
felt searching was useless.

But at least 2 people have complained that searching in the new scheme 
is also useless (returning too few results).

So two important principles to keep in mind:

1. It's not complainers you need to worry about; it's the people who 
never say anything.  The difference is that at least the complainers 
still care, whereas the latter just think it's not worth the effort.  
Silence is bad news.

2. You can't please everybody.

So we've got to figure out how to improve the issue searching (some 
discussion) keeping #2 in mind [so there are limits to how much 
perfection we can achieve].

Maybe just plug Google into our search?

Personally, I've trained myself to just "search" by using the search UI, 
plugging in a topic and looking for bugs marked not-resolved.  So that's 
probably a bad sign in that I probably have unconsciously absorbed the 
search field as being useless too.

What could be helpful (need-action) is for us to come up with search 
"test cases"

On Wed, Sep 8, 2010 at 2:32 AM, Eric Kow <bugs@darcs.net> wrote:
>
> Personally, I've trained myself to just "search" by using the search UI,
> plugging in a topic and looking for bugs marked not-resolved.  So that's
> probably a bad sign in that I probably have unconsciously absorbed the
> search field as being useless too.
>
> What could be helpful (need-action) is for us to come up with search
> "test cases"

Going straight to a ticket based on the issue number seems like one
reasonable use case.  I've become a fan of how trac handles this -- a
number prefixed by a # is treated as a specific ticket id look-up, not
a search.

The current behavior was particularly frustrating because, aside from
not finding anything for my search (which was an issue number), the
results list displayed one link -- a csv file containing the results
(nothing).  My reaction was to just click on the link in the results
list (I was expecting exactly one result, since I was under the
assumption that I had supplied sufficient information to unambiguously
specify a singular bug).  Now I have an empty csv file somewhere,
which isn't a problem per-say, but It also didn't help me find the bug
:).  Anyhow, it would be nice if no csv file was created for empty
result lists, but that may be a separate bug.

As an aside, I haven't been all that impressed with google-based
site-specific search engines.  It's very rare that the right
information is displayed in a search result snippet, particularly for
things like bug trackers.  Perhaps the results display can be
configured suitably, but I have yet to see an implementation that does
so.

--Rogan

>
> ----------
> messages: 12503
> nosy: Serware, creswick, dmitry.kurochkin, kowey
> priority: feature
> status: need-action
> title: reconsider searching from issue text instead of title
> topic: BugTracker
>
> __________________________________
> Darcs bug tracker <bugs@darcs.net>
> <http://bugs.darcs.net/issue1952>
> __________________________________
>