[mythtv] Re: [mythtv-commits] mythtv commits

[David Engel]

On Tue, Jun 01, 2004 at 07:16:42PM -0700, Bruce Markey wrote:
> I don't think about mythfilldatabase much 'cause I'm usually
> asleep at the time =). If it added significant time (even 30sec
> or a minute) but saved tenths of a second while the user was
> interacting it would be a win. If this is any indication, I
> don't think the text indexing will be a problem.

I ran some trial mythfilldatabase runs with and without FULLTEXT
indexes.  The worst run increase from 40 to 47 seconds.  Pretty
inconsequential, IMO.  I suspect the warning I read about slow inserts
only pertains to tables that are updated frequently.

> Fuzzy matches would be a good thing for the search pages but may
> cause false positives for k*Search rules, as you mentioned, and
> this could lead to some unintended conflicts.

Right.  That's why I felt the phrase matching was required.

> seconds. All the full text searches reported 0.00. However, it doesn't
> seem to support wildcards at all. Someone correct me if I'm wrong
> but you couldn't do "Junkyard%Wars" as you can with LIKE. Also,

Good point.

> it seems to do exact word matches as in this example with "John"
> and "Johnson"

I think you can do wildcards at the end of a work, e.g. 'John*'.

> "Vie" doesn't match? How big is their stopword list? In addition

I think they also use word length.  Too short and it won't get used.

> to each word being some form of exact match and no wildcards, each
> word is "or"d. This may not be what we need. For the purpose of
> setting automatic record rules, I kind of like wildcard or regex

Right.  That's why I would always use BOOLEAN MODE.  That allows you
to prefix words that must be present with '+' and words the must not
be present with '-', along with a few other useful operations.

At this point, if FULLTEXT gets added, I think it should probably be
enabled on a per search basis by using special syntax in the search
phrase.  For example, any search phrase starting with a '+' would
result in a BOOLEAN MODE FULLTEXT search.

David
-- 
David Engel
gigem at comcast.net