Search

When a control point offers a search box, it does not filter the list it already has. It sends the search to UMS as a UPnP criterion such as dc:title contains "beatles", and UMS answers with the matching items. How that criterion is turned into a query is what the settings on this page control.

The two search engines

UMS can answer a search in two ways:

• Full-text index (default). UMS keeps a Lucene index next to the database and searches that. It is fast even on large libraries and it is what makes partial words and typo tolerance possible.
• Plain database search. A straight SQL query. Slower on large libraries and it only matches what is literally there, but it has no index to go out of date.

The full-text index covers the file name of every media file, and for audio also the song title, album, artist, album artist, composer, conductor and genre.

UMS switches to the database search on its own when a control point sends a criterion the index cannot answer, so turning the index off is rarely necessary.

# Use the full-text index. FALSE falls back to a plain database search.
search_lucene_use_engine = TRUE

Searching for part of a word

Most search boxes send a contains criterion while you are still typing. UMS turns each word of your input into two alternatives at once:

• a prefix match, so beat already finds Beatles before you finish the word
• a fuzzy match, so beatlse still finds Beatles despite the typo

# Tolerate typos in addition to matching prefixes.
search_lucene_contains_fuzzy = TRUE

Both settings match prefixes. They differ in what else they match:

	TRUE (default)	FALSE
beat finds Beatles	yes	yes
beatlse finds Beatles	yes	no
unrelated results on short input	possible	no

The last row is the reason the setting exists. Typo tolerance allows up to two wrong, missing or extra letters, and on a short word that reaches surprisingly far: searching for ranz also returns everything called Jazz, because two substitutions get you from one to the other. If that noise bothers you more than the typo tolerance helps, set it to FALSE.

Two details worth knowing:

• A prefix is only used from three characters on. One or two letters would match a large part of the library without telling you anything the typo tolerance does not already find.
• Putting your input in quotation marks searches for the words next to each other instead, which is useful when the individual words are common but the combination is not.

Searching for an exact value

Some criteria ask for an exact value rather than a part of one — a control point browsing by genre sends upnp:genre = "Jazz", for instance. By default that is taken literally.

# Tolerate typos in exact matches as well. Off by default.
search_lucene_equal_fuzzy = FALSE

Turning this on makes exact comparisons tolerate typos too. That is rarely what you want, because these criteria usually come from the control point itself rather than from something you typed, and a value that is nearly right is then quietly treated as right.

Notes

Search ignores capitalization. chapitre, Chapitre and CHAPITRE return the same results.

If a search does not return what you expect, the debug log shows the query UMS built from your input, on a line beginning with lucene search string is. Comparing it with what you typed usually explains the result on the spot.