Documentation Source Text

Artifact Content
Login

Artifact 79ad85863f8932f46f5c8ee79a2fb7498b314f97167bddf3518e02596904f4f3:



<tcl>hd_keywords *fts5 FTS5</tcl>
<title>SQLite FTS5 Extension</title>

<table_of_contents>

<h1>Overview of FTS5</h1>

<p>FTS5 is an SQLite [virtual table module] that provides 
<a href=http://en.wikipedia.org/wiki/Full_text_search>full-text search</a>
functionality to database applications. In their most elementary form, 
full-text search engines allow the user to efficiently search a large 
collection of documents for the subset that contain one or more instances of a
search term. The search functionality provided to world wide web users by
<a href=www.google.com>Google</a> is, among other things, a full-text search
engine, as it allows users to search for all documents on the web that contain,
for example, the term "fts5".

<p>To use FTS5, the user creates an FTS5 virtual table with one or more
columns. For example:

<codeblock>
  CREATE VIRTUAL TABLE email USING fts5(sender, title, body);
</codeblock>

<p>It is an error to add types, constraints or [PRIMARY KEY] declarations to 
a CREATE VIRTUAL TABLE statement used to create an FTS5 table. Once created,
an FTS5 table may be populated using [INSERT], [UPDATE] or [DELETE] statements
like any other table. Like any other table with no PRIMARY KEY declaration, an
FTS5 table has an implicit INTEGER PRIMARY KEY field named rowid. 

<p>Not shown in the example above is that there are also 
[FTS5 CREATE TABLE Options | various options] that may be provided to FTS5 as
part of the CREATE VIRTUAL TABLE statement to configure various aspects of the
new table. These may be used to modify the way in which the FTS5 table extracts
terms from documents and queries, to create extra indexes on disk to speed up
prefix queries, or to create an FTS5 table that acts as an index on content
stored elsewhere.

<p>Once populated, there are three ways to execute a full-text query against
the contents of an FTS5 table:

<ul><li> Using a MATCH operator in the WHERE clause of a SELECT statement, or
    <li> Using an equals ("=") operator in the WHERE clause of a SELECT statement, or
    <li> using the [table-valued function] syntax.
</ul>

<p>If using the MATCH or = operators, the expression to the left of the MATCH
   operator is usually the name of the FTS5 table (the exception is when 
   [FTS5 column filters | specifying a column-filter]). The expression on the right
   must be a text value specifying the term to search for. For the table-valued
   function syntax, the term to search for is specified as the first table argument.
   For example:

<codeblock>
  <i>-- Query for all rows that contain at least once instance of the term</i>
  <i>-- "fts5" (in any column). The following three queries are equivalent.</i>
  SELECT * FROM email WHERE email MATCH 'fts5';
  SELECT * FROM email WHERE email = 'fts5';
  SELECT * FROM email('fts5');
</codeblock>

<p> By default, FTS5 full-text searches are case-independent. Like any other
SQL query that does not contain an ORDER BY clause, the example above returns
results in an arbitrary order. To sort results by relevance (most to least
relevant), an ORDER BY may be added to a full-text query as follows:

<codeblock>
  <i>-- Query for all rows that contain at least once instance of the term</i>
  <i>-- "fts5" (in any column). Return results in order from best to worst</i>
  <i>-- match.  </i>
  SELECT * FROM email WHERE email MATCH 'fts5' ORDER BY rank;
</codeblock>

<p> As well as the column values and rowid of a matching row, an application
may use [FTS5 auxiliary functions] to retrieve extra information regarding
the matched row. For example, an auxiliary function may be used to retrieve
a copy of a column value for a matched row with all instances of the matched
term surrounded by html &lt;b&gt;&lt;/b&gt; tags. Auxiliary functions are
invoked in the same way as SQLite [corefunc | scalar functions], except that the name
of the FTS5 table is specified as the first argument. For example:

<codeblock>
  <i>-- Query for rows that match "fts5". Return a copy of the "body" column</i>
  <i>-- of each row with the matches surrounded by &lt;b&gt;&lt;/b&gt; tags.</i>
  SELECT highlight(email, 2, '&lt;b&gt;', '&lt;/b&gt;') FROM email('fts5');
</codeblock>

<p>A description of the available auxiliary functions, and more details
regarding configuration of the special "rank" column, are 
[FTS5 auxiliary functions | available below]. [FTS5 custom auxiliary functions|
Custom auxiliary functions] may also be implemented in C and registered with
FTS5, just as custom SQL functions may be registered with the SQLite core.

<p> As well as searching for all rows that contain a term, FTS5 allows 
the user to search for rows that contain:

<ul>
  <li> any terms that begin with a specified prefix,
  <li> "phrases" - sequences of terms or prefix terms that must feature in a
       document for it to match the query, 
  <li> sets of terms, prefix terms or phrases that appear within a specified
       proximity of each other (these are called "NEAR queries"), or
  <li> boolean combinations of any of the above.
</ul>

<p> Such advanced searches are requested by providing a more complicated 
FTS5 query string as the text to the right of the MATCH operator (or =
operator, or as the first argument to a table-valued function syntax). The 
full query syntax is [FTS5 query syntax | described here].

<h1 tags="FTS5 building">Compiling and Using FTS5</h1>

<h2>Building FTS5 as part of SQLite</h2>

<p>As of [version 3.9.0] ([dateof:3.9.0]), 
FTS5 is included as part of the SQLite [amalgamation].
It is disabled by default. If using the two autoconf build system, it is
enabled by specifying the "--enable-fts5" option when running the configure
script. 

<p>Or, if sqlite3.c is compiled using some other build system, by arranging for
the SQLITE_ENABLE_FTS5 pre-processor symbol to be defined.

<h2>Building a Loadable Extension</h2>

<p>Alternatively, FTS5 may be built as a loadable extension.

<p>The canonical FTS5 source code consists of a series of *.c and other files
in the "ext/fts5" directory of the SQLite source tree. A build process reduces
this to just two files - "fts5.c" and "fts5.h" - which may be used to build an
SQLite loadable extension.

<ol>
  <li> Obtain the latest SQLite code from fossil.
  <li> Create a Makefile as described in [How To Compile SQLite].
  <li> Build the "fts5.c" target. Which also creates fts5.h.
</ol>

<codeblock>
  $ wget -c http://www.sqlite.org/src/tarball/SQLite-trunk.tgz?uuid=trunk -O SQLite-trunk.tgz
  .... output ...
  $ tar -xzf SQLite-trunk.tgz
  $ cd SQLite-trunk
  $ ./configure && make fts5.c
  ... lots of output ...
  $ ls fts5.&#91;ch]
  fts5.c        fts5.h
</codeblock>

<p>
  The code in "fts5.c" may then be compiled into a loadable extension or
  statically linked into an application as described in 
  [Compiling Loadable Extensions]. There are two entry points defined, both
  of which do the same thing:

<ul>
  <li> sqlite3_fts_init
  <li> sqlite3_fts5_init
</ul>

<p>
  The other file, "fts5.h", is not required to compile the FTS5 extension. 
  It is used by applications that implement [Extending FTS5 | custom FTS5 tokenizers or auxiliary functions].

<h1 tags="FTS5 query syntax">Full-text Query Syntax</h1>

<p>
The following block contains a summary of the FTS query syntax in BNF form.
A detailed explanation follows.

<codeblock>
&lt;phrase&gt;    := string &#91;*]
&lt;phrase&gt;    := &lt;phrase&gt; + &lt;phrase&gt;
&lt;neargroup&gt; := NEAR ( &lt;phrase&gt; &lt;phrase&gt; ... &#91;, N] )
&lt;query&gt;     := &#91; &#91;-] &lt;colspec&gt; :] &#91;&#94;] &lt;phrase&gt;
&lt;query&gt;     := &#91; &#91;-] &lt;colspec&gt; :] &lt;neargroup&gt;
&lt;query&gt;     := &#91; &#91;-] &lt;colspec&gt; :] ( &lt;query&gt; )
&lt;query&gt;     := &lt;query&gt; AND &lt;query&gt;
&lt;query&gt;     := &lt;query&gt; OR &lt;query&gt;
&lt;query&gt;     := &lt;query&gt; NOT &lt;query&gt;
&lt;colspec&gt;   := colname
&lt;colspec&gt;   := { colname1 colname2 ... }
</codeblock>

<h2 tags="FTS5 Strings">FTS5 Strings</h2>
<p>
Within an FTS expression a <b>string</b> may be specified in one of two ways:

<ul>
  <li> <p>By enclosing it in double quotes ("). Within a string, any embedded
       double quote characters may be escaped SQL-style - by adding a second
       double-quote character.

  <li> <p>As an FTS5 bareword that is not "AND", "OR" or "NOT" (case sensitive). 
       An FTS5 bareword is a string of one or more consecutive characters that
       are all either:
       
       <ul>
         <li> Non-ASCII range characters (i.e. unicode codepoints greater 
              than 127), or 
         <li> One of the 52 upper and lower case ASCII characters, or
         <li> One of the 10 decimal digit ASCII characters, or
         <li> The underscore character (unicode codepoint 96).
         <li> The substitute character (unicode codepoint 26).
       </ul>

       Strings that include any other characters must be quoted. Characters
       that are not currently allowed in barewords, are not quote characters and
       do not currently serve any special purpose in FTS5 query expressions may
       at some point in the future be allowed in barewords or used to implement
       new query functionality. This means that queries that are currently
       syntax errors because they include such a character outside of a quoted
       string may be interpreted differently by some future version of FTS5.
</ul>

<h2 tags="FTS5 Phrases">FTS5 Phrases</h2>
<p>
FTS queries are made up of <b>phrases</b>. A phrase is an ordered list of 
one or more tokens. A string is transformed into a phrase by passing it to
the FTS table tokenizer. Two phrases can be concatenated into a single 
large phrase using the "+" operator. For example, assuming the tokenizer
module being used tokenizes the input "one.two.three" to three separate
tokens, the following three queries all specify the same phrase:

<codeblock>
  ... MATCH '"one two three"'
  ... MATCH 'one + two + three'
  ... MATCH '"one two" + three'
  ... MATCH 'one.two.three'
</codeblock>

<p>
A phrase matches a document if the document contains at least one sub-sequence
of tokens that matches the sequence of tokens that make up the phrase.

<h2 tags="FTS5 prefix queries">FTS5 Prefix Queries</h2>
<p>
If a "*" character follows a string within an FTS expression, then the final
token extracted from the string is marked as a <b>prefix token</b>. As you
might expect, a prefix token matches any document token of which it is a 
prefix. For example, the first two queries in the following block will match
any document that contains the token "one" immediately followed by the token
"two" and then any token that begins with "thr".

<codeblock>
  ... MATCH '"one two thr" * '
  ... MATCH 'one + two + thr*'
  ... MATCH '"one two thr*"'      <b>-- May not work as expected!</b>
</codeblock>

<p>The final query in the block above may not work as expected. Because the
"*" character is inside the double-quotes, it will be passed to the tokenizer,
which will likely discard it (or perhaps, depending on the specific tokenizer
in use, include it as part of the final token) instead of recognizing it as
a special FTS character.

<tcl>hd_fragment carrotq {FTS5 initial token}</tcl>
<h2 tags="FTS5 initial token queries">FTS5 Initial Token Queries</h2>
<p>
If a "&#94;" character appears immediately before a phrase that is not part of a
NEAR query, then that phrase only matches a document only if it starts at the
first token in a column. The "&#94;" syntax may be combined with a 
[FTS5 column filters|column filter], but may not be inserted into the middle of
a phrase.

<codeblock>
  ... MATCH '&#94;one'              <i>-- first token in any column must be "one"</i>
  ... MATCH '&#94; one + two'       <i>-- phrase "one two" must appear at start of a column</i>
  ... MATCH '&#94; "one two"'       <i>-- same as previous </i>
  ... MATCH 'a : &#94;two'          <i>-- first token of column "a" must be "two"</i>
  ... MATCH 'NEAR(&#94;one, two)'   <b>-- syntax error! </b>
  ... MATCH 'one + &#94;two'        <b>-- syntax error! </b>
  ... MATCH '"&#94;one two"'        <b>-- May not work as expected!</b>
</codeblock>

<h2 tags="FTS5 NEAR queries">FTS5 NEAR Queries</h2>

<p>Two or more phrases may be grouped into a <b>NEAR group</b>. A NEAR group
is specified by the token "NEAR" (case sensitive) followed by an open
parenthesis character, followed by two or more whitespace separated phrases, optionally followed by a comma and the numeric parameter <i>N</i>, followed by
a close parenthesis. For example:

<codeblock>
  ... MATCH 'NEAR("one two" "three four", 10)'
  ... MATCH 'NEAR("one two" thr* + four)'
</codeblock>

<p>If no <i>N</i> parameter is supplied, it defaults to 10. A NEAR group
matches a document if the document contains at least one clump of tokens that: 

<ol> 
  <li> contains at least one instance of each phrase, and 
  <li> for which the number of tokens between the end of the first phrase 
       and the beginning of the last phrase in the clump is less than or equal to <i>N</i>.
</ol>

<p>For example:

<codeblock>
  CREATE VIRTUAL TABLE f USING fts5(x);
  INSERT INTO f(rowid, x) VALUES(1, 'A B C D x x x E F x');

  ... MATCH 'NEAR(e d, 4)';                      <i>-- Matches!</i>
  ... MATCH 'NEAR(e d, 3)';                      <i>-- Matches!</i>
  ... MATCH 'NEAR(e d, 2)';                      <i>-- Does not match!</i>

  ... MATCH 'NEAR("c d" "e f", 3)';              <i>-- Matches!</i>
  ... MATCH 'NEAR("c"   "e f", 3)';              <i>-- Does not match!</i>

  ... MATCH 'NEAR(a d e, 6)';                    <i>-- Matches!</i>
  ... MATCH 'NEAR(a d e, 5)';                    <i>-- Does not match!</i>

  ... MATCH 'NEAR("a b c d" "b c" "e f", 4)';    <i>-- Matches!</i>
  ... MATCH 'NEAR("a b c d" "b c" "e f", 3)';    <i>-- Does not match!</i>

</codeblock>

<h2 tags="FTS5 column filters">FTS5 Column Filters</h2>

<p>
A single phrase or NEAR group may be restricted to matching text within a
specified column of the FTS table by prefixing it with the column name 
followed by a colon character. Or to a set of columns by prefixing it
with a whitespace separated list of column names enclosed in parenthesis
("curly brackets") followed by a colon character. Column names may be specified
using either of the two forms described for strings above. Unlike strings that
are part of phrases, column names are not passed to the tokenizer module.
Column names are case-insensitive in the usual way for SQLite column names -
upper/lower case equivalence is understood for ASCII-range characters only.

<codeblock>
  ... MATCH 'colname : NEAR("one two" "three four", 10)'
  ... MATCH '"colname" : one + two + three'

  ... MATCH '{col1 col2} : NEAR("one two" "three four", 10)'
  ... MATCH '{col2 col1 col3} : one + two + three'
</codeblock>

<p>
If a column filter specification is preceded by a "-" character, then
it is interpreted as a list of column not to match against. For example:

<codeblock>
  <i>-- Search for matches in all columns except "colname"</i>
  ... MATCH '- colname : NEAR("one two" "three four", 10)'

  <i>-- Search for matches in all columns except "col1", "col2" and "col3"</i>
  ... MATCH '- {col2 col1 col3} : one + two + three'
</codeblock>

<p>
Column filter specifications may also be applied to arbitrary expressions
enclosed in parenthesis. In this case the column filter applies to all 
phrases within the expression. Nested column filter operations may only 
further restrict the subset of columns matched, they can not be used to 
re-enable filtered columns. For example:

<codeblock>
  <i>-- The following are equivalent:</i>
  ... MATCH '{a b} : ( {b c} : "hello" AND "world" )'
  ... MATCH '(b : "hello") AND ({a b} : "world")'
</codeblock>

<p>
Finally, a column filter for a single column may be specified by using
the column name as the LHS of a MATCH operator (instead of the usual
table name). For example:

<codeblock>
  <i>-- Given the following table</i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b, c);

  <i>-- The following are equivalent</i>
  SELECT * FROM ft WHERE b MATCH 'uvw AND xyz';
  SELECT * FROM ft WHERE ft MATCH 'b : (uvw AND xyz)';

  <i>-- This query cannot match any rows (since all columns are filtered out): </i>
  SELECT * FROM ft WHERE b MATCH 'a : xyz';
</codeblock>

<h2 tags="FTS5 boolean operators">FTS5 Boolean Operators</h2>

<p>
Phrases and NEAR groups may be arranged into expressions using <b>boolean
operators</b>. In order of precedence, from highest (tightest grouping) to
lowest (loosest grouping), the operators are:

<table striped=1>
  <tr><th>Operator <th>Function

  <tr><td><code>&lt;query1&gt; NOT &lt;query2&gt;</code> 
      <td>Matches if query1 matches and query2 does not match.

  <tr><td><code>&lt;query1&gt; AND &lt;query2&gt;</code> 
      <td>Matches if both query1 and query2 match.

  <tr><td><code>&lt;query1&gt; OR &lt;query2&gt;</code> 
      <td>Matches if either query1 or query2 match.

</table>

<p>
Parenthesis may be used to group expressions in order to modify operator
precedence in the usual ways. For example:

<codeblock>
  <i>-- Matches documents that contain at least one instance of either "one"</i>
  <i>-- or "two", but do not contain any instances of token "three".</i>
  ... MATCH 'one OR two NOT three'

  <i>-- Match all documents that contain the token "two" but not "three", or</i>
  <i>-- contain the token "one".</i>
  ... MATCH 'one OR (two NOT three)'
</codeblock>

<p>
Phrases and NEAR groups may also be connected by <b>implicit AND operators</b>.
For simplicity, these are not shown in the BNF grammar above. Essentially, any
sequence of phrases or NEAR groups (including those restricted to matching
specified columns) separated only by whitespace are handled as if there were an
implicit AND operator between each pair of phrases or NEAR groups. Implicit
AND operators are never inserted after or before an expression enclosed in
parenthesis. For example:

<codeblock>
  ... MATCH 'one two three'         <i>-- 'one AND two AND three'</i>
  ... MATCH 'three "one two"'       <i>-- 'three AND "one two"'</i>
  ... MATCH 'NEAR(one two) three'   <i>-- 'NEAR(one two) AND three'</i>
  ... MATCH 'one OR two three'      <i>-- 'one OR two AND three'</i>

  ... MATCH '(one OR two) three'    <i>-- Syntax error!</i>
  ... MATCH 'func(one two)'         <i>-- Syntax error!</i>
</codeblock>

<h1 tags="FTS5 CREATE TABLE Options">FTS5 Table Creation and Initialization</h1>

<p>Each argument specified as part of a "CREATE VIRTUAL TABLE ... USING fts5 
..." statement is either a column declaration or a configuration option. A
<b>column declaration</b> consists of one or more whitespace separated FTS5
barewords or string literals quoted in any manner acceptable to SQLite.

<p>The first string or bareword in a column declaration is the column name. It
is an error to attempt to name an fts5 table column "rowid" or "rank", or to
assign the same name to a column as is used by the table itself. This is not
supported.

<p>Each subsequent string or bareword in a column declaration is a column
option that modifies the behaviour of that column. Column options are
case-independent. Unlike the SQLite core, FTS5 considers unrecognized column
options to be errors. Currently, the only option recognized is 
[unindexed | "UNINDEXED" (see below)].

<p>A <b>configuration option</b> consists of an FTS5 bareword - the option name -
followed by an "=" character, followed by the option value. The option value is
specified using either a single FTS5 bareword or a string literal, again quoted
in any manner acceptable to the SQLite core. For example:

<codeblock>
  CREATE VIRTUAL TABLE mail USING fts5(sender, title, body, tokenize = 'porter ascii');
</codeblock>

<p> There are currently the following configuration options:

<ul>
  <li> The "tokenize" option, used to configure a [FTS5 tokenizers | custom tokenizer].
  <li> The "prefix" option, used to add [FTS5 prefix indexes | prefix indexes]
       to an FTS5 table.
  <li> The "content" option, used to make the FTS5 table an 
       [FTS5 content option | external content or contentless table].
  <li> The "content_rowid" option, used to set the rowid field of an 
       [FTS5 external content tables | external content table].
  <li> The [FTS5 columnsize option | "columnsize" option], used to configure
       whether or not the size in tokens of each value in the FTS5 table is
       stored separately within the database.
  <li> The [FTS5 detail option | "detail" option]. This option may be used 
       to reduce the size of the FTS index on disk by omitting some information
       from it.  
</ul>

<h2 tags="unindexed">The UNINDEXED column option</h2>

<p>The contents of columns qualified with the UNINDEXED column option are not
added to the FTS index. This means that for the purposes of MATCH queries and
[FTS5 auxiliary functions], the column contains no matchable tokens. 

<p>For example, to avoid adding the contents of the "uuid" field to the FTS
index:
<codeblock>
  CREATE VIRTUAL TABLE customers USING fts5(name, addr, uuid UNINDEXED);
</codeblock>

<h2 tags="FTS5 prefix indexes">Prefix Indexes</h2>

<p> By default, FTS5 maintains a single index recording the location of each
token instance within the document set. This means that querying for complete
tokens is fast, as it requires a single lookup, but querying for a prefix 
token can be slow, as it requires a range scan. For example, to query for
the prefix token "abc*" requires a range scan of all tokens greater than
or equal to "abc" and less than "abd".

<p> A prefix index is a separate index that records the location of all
instances of prefix tokens of a certain length in characters used to speed
up queries for prefix tokens. For example, optimizing a query for prefix
token "abc*" requires a prefix index of three-character prefixes.

<p> To add prefix indexes to an FTS5 table, the "prefix" option is set to
either a single positive integer or a text value containing a white-space
separated list of one or more positive integer values. A prefix index is
created for each integer specified. If more than one "prefix" option is
specified as part of a single CREATE VIRTUAL TABLE statement, all apply.

<codeblock>
  <i>-- Two ways to create an FTS5 table that maintains prefix indexes for
  -- two and three character prefix tokens.</i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b, prefix='2 3');
  CREATE VIRTUAL TABLE ft USING fts5(a, b, prefix=2, prefix=3);
</codeblock>

<h2 tags="FTS5 tokenizers">Tokenizers</h2>

<p> The CREATE VIRTUAL TABLE "tokenize" option is used to configure the
specific tokenizer used by the FTS5 table. The option argument must be either
an FTS5 bareword, or an SQL text literal. The text of the argument is itself
treated as a white-space series of one or more FTS5 barewords or SQL text
literals. The first of these is the name of the tokenizer to use. The second
and subsequent list elements, if they exist, are arguments passed to the
tokenizer implementation.

<p> Unlike option values and column names, SQL text literals intended as
tokenizers must be quoted using single quote characters. For example:

<codeblock>
  <i>-- The following are all equivalent</i>
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = 'porter ascii');
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = "porter ascii");
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = "'porter' 'ascii'");
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = '''porter'' ''ascii''');

  <i>-- But this will fail:</i>
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = '"porter" "ascii"');

  <i>-- This will fail too:</i>
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = 'porter' 'ascii');
</codeblock>


<p>
FTS5 features three built-in tokenizer modules, described in subsequent
sections:

<ul>
  <li> The <b>unicode61</b> tokenizer, based on the Unicode 6.1 standard. This
       is the default.

  <li> The <b>ascii</b> tokenizer, which assumes all characters outside of
  the ASCII codepoint range (0-127) are to be treated as token characters.

  <li> The <b>porter</b> tokenizer, which implements the 
<a href=http://tartarus.org/martin/PorterStemmer/>porter stemming algorithm</a>.
</ul>

<p> It is also possible to create custom tokenizers for FTS5. The API for doing so is [custom tokenizers | described here].

<h3>Unicode61 Tokenizer</h3>

<p> The unicode tokenizer classifies all unicode characters as either 
"separator" or "token" characters. By default all space and punctuation
characters, as defined by Unicode 6.1, are considered separators, and all 
other characters as token characters. More specifically, all unicode 
characters assigned to a 
<a href=https://en.wikipedia.org/wiki/Unicode_character_property#General_Category>
general category</a> beginning with "L" or "N" (letters and numbers,
specifically) or to category "Co" ("other, private use") are considered tokens.
All other characters are separators.
 
<p>Each contiguous run of one or more token characters is considered to be a
token. The tokenizer is case-insensitive according to the rules defined by
Unicode 6.1.

<p> By default, diacritics are removed from all Latin script characters. This
means, for example, that "A", "a", "&#192;", "&#224;", "&#194;" and "&#226;"
are all considered to be equivalent.

<p> Any arguments following "unicode61" in the token specification are treated
as a list of alternating option names and values. Unicode61 supports the
following options:

<table striped=1>
  <tr><th> Option <th> Usage
  <tr><td> remove_diacritics
  <td>This option should be set to "0" or "1". If it is set (the default),
  diacritics are removed from all latin script characters as described above.
  If it is clear, they are not. 

  <tr><td> categories
  <td>This option may be used to modify the set of Unicode general categories
  that are considered to correspond to token characters. The argument must
  consist of a space separated list of two-character general category
  abbreviations (e.g. "Lu" or "Nd"), or of the same with the second character
  replaced with an asterix ("*"), interpreted as a glob pattern. The default
  value is "L* N* Co".

  <tr><td> tokenchars
  <td> This option is used to specify additional unicode characters that 
  should be considered token characters, even if they are white-space or
  punctuation characters according to Unicode 6.1. All characters in the
  string that this option is set to are considered token characters.

  <tr><td> separators
  <td> This option is used to specify additional unicode characters that 
  should be considered as separator characters, even if they are token
  characters according to Unicode 6.1. All characters in the string that 
  this option is set to are considered separators.
</table>

<p> For example:

<codeblock>
  <i>-- Create an FTS5 table that does not remove diacritics from Latin
  -- script characters, and that considers hyphens and underscore characters
  -- to be part of tokens. </i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b, 
      tokenize = "unicode61 remove_diacritics 0 tokenchars '-_'"
  );
</codeblock>

<p> or:

<codeblock>
  <i>-- Create an FTS5 table that, as well as the default token character classes,</i>
  <i>-- considers characters in class "Mn" to be token characters.</i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b, 
      tokenize = "unicode61 categories 'L* N* Co Mn'"
  );
</codeblock>

<p> The fts5 unicode61 tokenizer is byte-for-byte compatible with the fts3/4
unicode61 tokenizer.

<h3>Ascii Tokenizer</h3>

<p> The Ascii tokenizer is similar to the Unicode61 tokenizer, except that:

<ul>
  <li> All non-ASCII characters (those with codepoints greater than 127) are
  always considered token characters. If any non-ASCII characters are specified
  as part of the separators option, they are ignored.  

  <li> Case-folding is only performed for ASCII characters. So while "A" and
  "a" are considered to be equivalent, "&#195" and "&#227;" are distinct.

  <li> The remove_diacritics option is not supported.
</ul>

<p> For example:

<codeblock>
  <i>-- Create an FTS5 table that uses the ascii tokenizer, but does not
  -- consider numeric characters to be part of tokens.</i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b, 
      tokenize = "ascii separators '0123456789'"
  );
</codeblock>

<h3>Porter Tokenizer</h3>

<p> The porter tokenizer is a wrapper tokenizer. It takes the output of some
other tokenizer and applies the 
<a href=http://tartarus.org/martin/PorterStemmer/>porter stemming algorithm</a>
to each token before it returns it to FTS5. This allows search terms like
"correction" to match similar words such as "corrected" or "correcting". The
porter stemmer algorithm is designed for use with English language terms 
only - using it with other languages may or may not improve search utility.

<p> By default, the porter tokenizer operates as a wrapper around the default
tokenizer (unicode61). Or, if one or more extra arguments are added to the
"tokenize" option following "porter", they are treated as a specification for
the underlying tokenizer that the porter stemmer uses. For example:

<codeblock>
  <i>-- Two ways to create an FTS5 table that uses the porter tokenizer to
  -- stem the output of the default tokenizer (unicode61). </i>
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = porter); 
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = 'porter unicode61');

  <i>-- A porter tokenizer used to stem the output of the unicode61 tokenizer,
  -- with diacritics removed before stemming.</i>
  CREATE VIRTUAL TABLE t1 USING fts5(x, tokenize = 'porter unicode61 remove_diacritics 1');
</codeblock>

<h2 tags="FTS5 content option">External Content and Contentless Tables</h2>

<p>
Normally, when a row is inserted into an FTS5 table, as well as the various
full-text index entries and other data a copy of the row is stored in a private
table managed by the FTS5 module. When column values are requested from the
FTS5 table by the user or by an auxiliary function implementation, they are
read from this private table. The "content" option may be used to create an
FTS5 table that stores only FTS full-text index entries. Because the column
values themselves are usually much larger than the associated full-text index
entries, this can save significant database space.

<p>
There are two ways to use the "content" option:
<ul>
  <li> By setting it to an empty string to create a contentless FTS5 table. In
       this case FTS5 assumes that the original column values are unavailable
       to it when processing queries. Full-text queries and some auxiliary
       functions can still be used, but no column values apart from the rowid
       may be read from the table.

  <li> By setting it to the name of a database object (table, virtual table or
       view) that may be queried by FTS5 at any time to retrieve the column
       values. This is known as an "external content" table. In this case all
       FTS5 functionality may be used, but it is the responsibility of the user
       to ensure that the contents of the full-text index are consistent with
       the named database object. If they are not, query results may be
       unpredictable.  
</ul>

<h3 tags="FTS5 contentless tables">Contentless Tables</h3>

<p> A contentless FTS5 table is created by setting the "content" option to
an empty string. For example:

<codeblock>
  CREATE VIRTUAL TABLE f1 USING fts5(a, b, c, content='');
</codeblock>

<p> Contentless FTS5 tables do not support UPDATE or DELETE statements, or
INSERT statements that do not supply a non-NULL value for the rowid field.
Contentless tables do not support REPLACE conflict handling. REPLACE 
and INSERT OR REPLACE statements are treated as regular INSERT statements.
Rows may be deleted from a contentless table using an [FTS5 delete command].

<p> Attempting to read any column value except the rowid from a contentless
FTS5 table returns an SQL NULL value.

<h3 tags="FTS5 external content tables">External Content Tables</h3>

<p> An external content FTS5 table is created by setting the content 
option to the name of a table, virtual table or view (hereafter the "content
table") within the same database. Whenever column values are required by
FTS5, it queries the content table as follows, with the rowid of the row
for which values are required bound to the SQL variable:

<codeblock>
  SELECT &lt;content_rowid&gt;, &lt;cols&gt; FROM &lt;content&gt; WHERE &lt;content_rowid&gt; = ?;
</codeblock>

<p> In the above, &lt;content&gt; is replaced by the name of the content table.
By default, &lt;content_rowid&gt; is replaced by the literal text "rowid". Or,
if the "content_rowid" option is set within the CREATE VIRTUAL TABLE statement,
by the value of that option. &lt;cols&gt; is replaced by a comma-separated list
of the FTS5 table column names. For example:

<codeblock>
  <i>-- If the database schema is: </i>
  CREATE TABLE tbl (a, b, c, d INTEGER PRIMARY KEY);
  CREATE VIRTUAL TABLE fts USING fts5(a, c, content=tbl, content_rowid=d);

  <i>-- Fts5 may issue queries such as:</i>
  SELECT d, a, c FROM tbl WHERE d = ?;
</codeblock>

<p> The content table may also be queried as follows:

<codeblock>
  SELECT &lt;content_rowid&gt;, &lt;cols&gt; FROM &lt;content&gt; ORDER BY &lt;content_rowid&gt; ASC;
  SELECT &lt;content_rowid&gt;, &lt;cols&gt; FROM &lt;content&gt; ORDER BY &lt;content_rowid&gt; DESC;
</codeblock>

<p> It is still the responsibility of the user to ensure that the contents of
an external content FTS5 table are kept up to date with the content table. 
One way to do this is with triggers. For example:

<codeblock>
  <i>-- Create a table. And an external content fts5 table to index it.</i>
  CREATE TABLE tbl(a INTEGER PRIMARY KEY, b, c);
  CREATE VIRTUAL TABLE fts_idx USING fts5(b, c, content='tbl', content_rowid='a');

  <i>-- Triggers to keep the FTS index up to date.</i>
  CREATE TRIGGER tbl_ai AFTER INSERT ON tbl BEGIN
    INSERT INTO fts_idx(rowid, b, c) VALUES (new.a, new.b, new.c);
  END;
  CREATE TRIGGER tbl_ad AFTER DELETE ON tbl BEGIN
    INSERT INTO fts_idx(fts_idx, rowid, b, c) VALUES('delete', old.a, old.b, old.c);
  END;
  CREATE TRIGGER tbl_au AFTER UPDATE ON tbl BEGIN
    INSERT INTO fts_idx(fts_idx, rowid, b, c) VALUES('delete', old.a, old.b, old.c);
    INSERT INTO fts_idx(rowid, b, c) VALUES (new.a, new.b, new.c);
  END;
</codeblock>

<p> Like contentless tables, external content tables do not support REPLACE
conflict handling. Any operations that specify REPLACE conflict handling are
handled using ABORT.

<h2 tags="FTS5 columnsize option">The Columnsize Option</h2>

<p>Normally, FTS5 maintains a special backing table within the database that
stores the size of each column value in tokens inserted into the main FTS5
table in a separate table. This backing table is used by the
<a href=#xColumnSize>xColumnSize<a> API function, which is in turn used by
the built-in [FTS5 bm25 | bm25 ranking function] (and is likely to be useful 
to other ranking functions as well).

<p>In order to save space, this backing table may be omitted by setting the
columnsize option to zero. For example:

<codeblock>
  <i>-- A table without the xColumnSize() values stored on disk:</i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b, c, columnsize=0);

  <i>-- Three equivalent ways of creating a table that does store the</i>
  <i>-- xColumnSize() values on disk:</i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b, c);
  CREATE VIRTUAL TABLE ft USING fts5(a, b, c, columnsize=1);
  CREATE VIRTUAL TABLE ft USING fts5(a, b, columnsize='1', c);
</codeblock>

<p> It is an error to set the columnsize option to any value other than
0 or 1.

<p> If an FTS5 table is configured with columnsize=0 but is not a
[FTS5 contentless tables | contentless table], the xColumnSize API function
still works, but runs much more slowly. In this case, instead of reading
the value to return directly from the database, it reads the text value
itself and count the tokens within it on demand.

<p>Or, if the table is also a [FTS5 contentless tables | contentless table],
then the following apply:

<ul>
  <li> <p>The xColumnSize API always returns -1. There is no way to determine 
       the number of tokens in a value stored within a contentless FTS5 table
       configured with columnsize=0.

  <li> <p>Each inserted row must be accompanied by an explicitly specified rowid
       value. If a contentless table is configured with columnsize=0,
       attempting to insert a NULL value into the rowid is an SQLITE_MISMATCH
       error.

  <li> <p>All queries on the table must be full-text queries. In other words,
       they must use the MATCH or = operator with the table-name column as the
       left-hand operand, or else use the table-valued function syntax. Any
       query that is not a full-text query results in an error.
</ul>

<p> The name of the table in which the xColumnSize values are stored
(unless columnsize=0 is specified) is "&lt;name&gt;_docsize", where 
&lt;name&gt; is the name of the FTS5 table itself. The 
<a href=https://www.sqlite.org/download.html>sqlite3_analyzer</a>
tool may be used on an existing database in order to determine how much
space might be saved by recreating an FTS5 table using columnsize=0.

<h2 tags="FTS5 detail option">The Detail Option</h2>

<p> For each term in a document, the FTS index maintained by FTS5 
stores the rowid of the document, the column number of the column that contains
the term and the offset of the term within the column value. The "detail"
option may be used to omit some of this information. This reduces the space
that the index consumes within the database file, but also reduces the
capability and efficiency of the system.

<p> The detail option may be set to "full" (the default value), "column" or
"none". For example:

<codeblock>
  <i>-- The following two lines are equivalent (because the default value</i>
  <i>-- of "detail" is "full". </i>
  CREATE VIRTUAL TABLE ft1 USING fts5(a, b, c);
  CREATE VIRTUAL TABLE ft1 USING fts5(a, b, c, detail=full);

  CREATE VIRTUAL TABLE ft2 USING fts5(a, b, c, detail=column);
  CREATE VIRTUAL TABLE ft3 USING fts5(a, b, c, detail=none);
</codeblock>

<p>If the detail option is set to <b>column</b>, then for each term the FTS
index records the rowid and column number only, omitting the term offset
information. This results in the following restrictions:

<ul>
  <li> NEAR queries are not available.
  <li> Phrase queries are not available.
  <li> Assuming the table is not also a 
  [FTS5 contentless tables | contentless table], the 
  <a href=#xInstCount>xInstCount</a>, <a href=#xInst>xInst</a>, 
  <a href=#xPhraseFirst>xPhraseFirst</a> and <a href=#xPhraseNext>xPhraseNext</a>
  are slower than usual. This is because instead of reading the required data
  directly from the FTS index they have to load and tokenize the document text 
  on demand.
  <li> If the table is also a contentless table, the xInstCount, xInst, 
  xPhraseFirst and xPhraseNext APIs behave as if the current row contains no
  phrase matches at all (i.e. xInstCount() returns 0).
</ul>
  
<p>If the detail option is set to <b>none</b>, then for each term the FTS
index records just the rowid is stored. Both column and offset information
are omitted. As well as the restrictions itemized above for detail=column
mode, this imposes the following extra limitations:

<ul>
  <li> Column filter queries are not available.
  <li> Assuming the table is not also a contentless table, the 
  <a href=#xPhraseFirstColumn>xPhraseFirstColumn</a> and 
  <a href=#xPhraseNextColumn>xPhraseNextColumn</a> are slower than usual. 

  <li> If the table is also a contentless table, the xPhraseFirstColumn and
  xPhraseNextColumn APIs behave as if the current row contains no phrase
  matches at all (i.e. xPhraseFirstColumn() sets the iterator to EOF).
</ul>

<p> In one test that indexed a large set of emails (1636 MiB on disk), the FTS
index was 743 MiB on disk with detail=full, 340 MiB with detail=column and 134
MiB with detail=none.

<h1 tags="FTS5 auxiliary functions"> Auxiliary Functions </h1>

<p> Auxiliary functions are similar to [corefunc | SQL scalar functions],
except that they may only be used within full-text queries (those that use
the MATCH operator) on an FTS5 table. Their results are calculated based not
only on the arguments passed to them, but also on the current match and 
matched row. For example, an auxiliary function may return a numeric value
indicating the accuracy of the match (see the [FTS5 bm25| bm25()] function), 
or a fragment of text from the matched row that contains one or more 
instances of the search terms (see the [FTS5 snippet | snippet()] function).

<p>To invoke an auxiliary function, the name of the FTS5 table should be
specified as the first argument. Other arguments may follow the first,
depending on the specific auxiliary function being invoked. For example, to
invoke the "highlight" function:

<codeblock>
  SELECT highlight(email, 2, '&lt;b&gt;', '&lt;/b&gt;') FROM email WHERE email MATCH 'fts5'
</codeblock>

<p>The built-in auxiliary functions provided as part of FTS5 are described in
the following section. Applications may also implement 
[FTS5 custom auxiliary functions | custom auxiliary functions in C].

<h2>Built-in Auxiliary Functions</h2>

<p> FTS5 provides three built-in auxiliary functions:

<ul>
  <li> The [FTS5 bm25 | bm25() auxiliary function] returns a real value
       reflecting the accuracy of the current match. Better matches are
       assigned numerically lower values.

  <li> The [FTS5 highlight | highlight() auxiliary function] returns a copy
       of the text from one of the columns of the current match with each
       instance of a queried term within the result surrounded by specified
       markup (for example "&lt;b&gt;" and "&lt;/b&gt;").

  <li> The [FTS5 snippet | snippet() auxiliary function] selects a short
       fragment of text from one of the columns of the matched row and returns
       it with each instance of a queried term surrounded by markup in
       the same manner as the highlight() function. The fragment of text is
       selected so as to maximize the number of queried terms it contains.
</ul>

<h3 tags="FTS5 bm25">The bm25() function</h3>

<p> The built-in auxiliary function bm25() returns a real value indicating
how well the current row matches the full-text query. The better the match,
the numerically smaller the value returned. A query such as the following may
be used to return matches in order from best to worst match:

<codeblock>
  SELECT * FROM fts WHERE fts MATCH ? ORDER BY bm25(fts)
</codeblock>

<p> In order to calculate a documents score, the full-text query is separated
    into its component phrases. The bm25 score for document <i>D</i> and 
    query <i>Q</i> is then calculated as follows:

<p> <img src="images/fts5_formula1.png" style="width:75ex;margin-left:5ex">

<p> In the above, <i>nPhrase</i> is the number of phrases in the query.
    <i>|D|</i> is the number of tokens in the current document, and
    <i>avgdl</i> is the average number of tokens in all documents within the
    FTS5 table.  <i>k<sub>1</sub></i> and <i>b</i> are both constants,
    hard-coded at 1.2 and 0.75 respectively.

<p> The "-1" term at the start of the formula is not found in most
implementations of the BM25 algorithm. Without it, a better match is assigned
a numerically higher BM25 score. Since the default sorting order is
"ascending", this means that appending "ORDER BY bm25(fts)" to a query would
cause results to be returned in order from worst to best. The "DESC" keyword
would be required in order to return the best matches first. In order to
avoid this pitfall, the FTS5 implementation of BM25 multiplies the result
by -1 before returning it, ensuring that better matches are assigned
numerically lower scores.

<p> <i>IDF(q<sub>i</sub>)</i> is the inverse-document-frequency of query 
    phrase <i>i</i>. It is calculated as follows, where <i>N</i> is the total
    number of rows in the FTS5 table and <i>n(q<sub>i</sub>)</i> is the total
    number of rows that contain at least one instance of phrase <i>i</i>:

<p> <img src="images/fts5_formula2.png" style="width:75ex;margin-left:5ex">

<p> Finally, <i>f(q<sub>i</sub>,D)</i> is the phrase frequency of phrase 
<i>i</i>. By default, this is simply the number of occurrences of the phrase
within the current row. However, by passing extra real value arguments to 
the bm25() SQL function, each column of the table may be assigned a different
weight and the phrase frequency calculated as follows:

<p> <img src="images/fts5_formula3.png" style="width:75ex;margin-left:5ex">

<p> where <i>w<sub>c</sub></i> is the weight assigned to column <i>c</i> and
<i>n(q<sub>i</sub>,c)</i> is the number of occurrences of phrase <i>i</i> in
column <i>c</i> of the current row. The first argument passed to bm25()
following the table name is the weight assigned to the leftmost column of
the FTS5 table. The second is the weight assigned to the second leftmost
column, and so on. If there are not enough arguments for all table columns,
remaining columns are assigned a weight of 1.0. If there are too many 
trailing arguments, the extras are ignored. For example:

<codeblock>
  <i>-- Assuming the following schema:</i>
  CREATE VIRTUAL TABLE email USING fts5(sender, title, body);

  <i>-- Return results in bm25 order, with each phrase hit in the "sender"</i>
  <i>-- column considered the equal of 10 hits in the "body" column, and</i>
  <i>-- each hit in the "title" column considered as valuable as 5 hits in</i>
  <i>-- the "body" column.</i>
  SELECT * FROM email WHERE email MATCH ? ORDER BY bm25(email, 10.0, 5.0);
</codeblock>

<p>Refer to wikipedia for 
<a href="http://en.wikipedia.org/wiki/Okapi_BM25">more information regarding
BM25</a> and its variants.

<h3 tags="FTS5 highlight">The highlight() function</h3>

<p> The highlight() function returns a copy of the text from a specified 
column of the current row with extra markup text inserted to mark the start 
and end of phrase matches. 

<p>The highlight() must be invoked with exactly three arguments following 
the table name. To be interpreted as follows:

<ol>
  <li> An integer indicating the index of the FTS table column to read the 
       text from. Columns are numbered from left to right starting at zero.

  <li> The text to insert before each phrase match.

  <li> The text to insert after each phrase match.
</ol>

<p>For example:

<codeblock>
  <i>-- Return a copy of the text from the leftmost column of the current</i>
  <i>-- row, with phrase matches marked using html "b" tags.</i>
  SELECT highlight(fts, 0, '&lt;b&gt;', '&lt;/b&gt;') FROM fts WHERE fts MATCH ?
</codeblock>

<p>In cases where two or more phrase instances overlap (share one or more
tokens in common), a single open and close marker is inserted for each set
of overlapping phrases. For example:

<codeblock>
  <i>-- Assuming this:</i>
  CREATE VIRTUAL TABLE ft USING fts5(a);
  INSERT INTO ft VALUES('a b c x c d e');
  INSERT INTO ft VALUES('a b c c d e');
  INSERT INTO ft VALUES('a b c d e');

  <i>-- The following SELECT statement returns these three rows:</i>
  <i>--   '&#91;a b c&#93; x &#91;c d e&#93;'</i>
  <i>--   '&#91;a b c&#93; &#91;c d e&#93;'</i>
  <i>--   '&#91;a b c d e&#93;'</i>
  SELECT highlight(ft, 0, '&#91;', '&#93;') FROM ft WHERE ft MATCH 'a+b+c AND c+d+e';
</codeblock>

<h3 tags="FTS5 snippet">The snippet() function</h3>

<p>The snippet() function is similar to highlight(), except that instead of
returning entire column values, it automatically selects and extracts a
short fragment of document text to process and return. The snippet() function
must be passed five parameters following the table name argument:

<ol>
  <li> An integer indicating the index of the FTS table column to select
       the returned text from. Columns are numbered from left to right 
       starting at zero. A negative value indicates that the column should
       be automatically selected.

  <li> The text to insert before each phrase match within the returned text.

  <li> The text to insert after each phrase match within the returned text.

  <li> The text to add to the start or end of the selected text to indicate
       that the returned text does not occur at the start or end of its column,
       respectively.

  <li> The maximum number of tokens in the returned text. This must be greater
       than zero and equal to or less than 64. 
</ol>

<h2 tags="auxiliary function mapping">Sorting by Auxiliary Function Results</h2>

<p> All FTS5 tables feature a special hidden column named "rank". If the
current query is not a full-text query (i.e. if it does not include a MATCH
operator), the value of the "rank" column is always NULL. Otherwise, in a
full-text query, column rank contains by default the same value as would be
returned by executing the bm25() auxiliary function with no trailing 
arguments.

<p> The difference between reading from the rank column and using the bm25()
function directly within the query is only significant when sorting by the
returned value. In this case, using "rank" is faster than using bm25().

<codeblock>
  <i>-- The following queries are logically equivalent. But the second may</i>
  <i>-- be faster, particularly if the caller abandons the query before</i>
  <i>-- all rows have been returned (or if the queries were modified to </i>
  <i>-- include LIMIT clauses).</i>
  SELECT * FROM fts WHERE fts MATCH ? ORDER BY bm25(fts);
  SELECT * FROM fts WHERE fts MATCH ? ORDER BY rank;
</codeblock>

<p> Instead of using bm25() with no trailing arguments, the specific auxiliary
function mapped to the rank column may be configured either on a per-query
basis, or by setting a different persistent default for the FTS table.

<p> In order to change the mapping of the rank column for a single query, 
a term similar to either of the following is added to the WHERE clause of a 
query:

<codeblock>
  rank MATCH 'auxiliary-function-name(arg1, arg2, ...)'
  rank = 'auxiliary-function-name(arg1, arg2, ...)'
</codeblock>

<p> The right-hand-side of the MATCH or = operator must be a constant
expression that evaluates to a string consisting of the auxiliary function to
invoke, followed by zero or more comma separated arguments within parenthesis.
Arguments must be SQL literals. For example:

<codeblock>
  <i>-- The following queries are logically equivalent. But the second may</i>
  <i>-- be faster. See above. </i>
  SELECT * FROM fts WHERE fts MATCH ? ORDER BY bm25(fts, 10.0, 5.0);
  SELECT * FROM fts WHERE fts MATCH ? AND rank MATCH 'bm25(10.0, 5.0)' ORDER BY rank;
</codeblock>

<p> The table-valued function syntax may also be used to specify an alternative
ranking function. In this case the text describing the ranking function should
be specified as the second table-valued function argument. The following three
queries are equivalent:

<codeblock>
  SELECT * FROM fts WHERE fts MATCH ? AND rank MATCH 'bm25(10.0, 5.0)' ORDER BY rank;
  SELECT * FROM fts WHERE fts = ? AND rank = 'bm25(10.0, 5.0)' ORDER BY rank;
  SELECT * FROM fts WHERE fts(?, 'bm25(10.0, 5.0)') ORDER BY rank;
</codeblock>

<p> The default mapping of the rank column for a table may be modified 
using the [FTS5 rank configuration option].

<h1>Special INSERT Commands</h1>

<h2 tags="FTS5 automerge option">The 'automerge' Configuration Option</h2>

<p>
  Instead of using a single data structure on disk to store the full-text
  index, FTS5 uses a series of b-trees. Each time a new transaction is
  committed, a new b-tree containing the contents of the committed transaction
  is written into the database file. When the full-text index is queried, each
  b-tree must be queried individually and the results merged before being
  returned to the user.

<p>
  In order to prevent the number of b-trees in the database from becoming too
  large (slowing down queries), smaller b-trees are periodically merged into
  single larger b-trees containing the same data. By default, this happens
  automatically within INSERT, UPDATE or DELETE statements that modify the
  full-text index. The 'automerge' parameter determines how many smaller
  b-trees are merged together at a time. Setting it to a small value can
  speed up queries (as they have to query and merge the results from fewer 
  b-trees), but can also slow down writing to the database (as each INSERT,
  UPDATE or DELETE statement has to do more work as part of the automatic
  merging process).

<p>
  Each of the b-trees that make up the full-text index is assigned to a "level"
  based on its size. Level-0 b-trees are the smallest, as they contain the
  contents of a single transaction. Higher level b-trees are the result of
  merging two or more level-0 b-trees together and so they are larger. FTS5
  begins to merge b-trees together once there exist <i>M</i> or more b-trees 
  with the same level, where <i>M</i> is the value of the 'automerge' 
  parameter.

<p>
  The maximum allowed value for the 'automerge' parameter is 16. The default
  value is 4. Setting the 'automerge' parameter to 0 disables the automatic 
  incremental merging of b-trees altogether.

<codeblock>
  INSERT INTO ft(ft, rank) VALUES('automerge', 8);
</codeblock>

<h2>The 'crisismerge' Configuration Option</h2>

<p>The 'crisismerge' option is similar to 'automerge', in that it determines
how and how often the component b-trees that make up the full-text index are
merged together. Once there exist <i>C</i> or more b-trees on a single level
within the full-text index, where <i>C</i> is the value of the 'crisismerge'
option, all b-trees on the level are immediately merged into a single b-tree.

<p>The difference between this option and the 'automerge' option is that when
the 'automerge' limit is reached FTS5 only begins to merge the b-trees
together. Most of the work is performed as part of subsequent INSERT, 
UPDATE or DELETE operations. Whereas when the 'crisismerge' limit is reached,
the offending b-trees are all merged immediately. This means that an INSERT,
UPDATE or DELETE that triggers a crisis-merge may take a long time to 
complete.

<p>The default 'crisismerge' value is 16. There is no maximum limit. Attempting
to set the 'crisismerge' parameter to a value of 0 or 1 is equivalent to
setting it to the default value (16). It is an error to attempt to set the
'crisismerge' option to a negative value.

<codeblock>
  INSERT INTO ft(ft, rank) VALUES('crisismerge', 16);
</codeblock>

<h2 tags="FTS5 delete command">The 'delete' Command</h2>

<p> This command is only available with [FTS5 external content tables |
external content] and [FTS5 contentless tables | contentless] tables. It
is used to delete the index entries associated with a single row from the
full-text index. This command and the [FTS5 delete-all command | delete-all]
command are the only ways to remove entries from the full-text index of a
contentless table.

<p> In order to use this command to delete a row, the text value 'delete' 
must be inserted into the special column with the same name as the table.
The rowid of the row to delete is inserted into the rowid column. The
values inserted into the other columns must match the values currently
stored in the table. For example:

<codeblock>
  <i>-- Insert a row with rowid=14 into the fts5 table.</i>
  INSERT INTO ft(rowid, a, b, c) VALUES(14, $a, $b, $c);
  
  <i>-- Remove the same row from the fts5 table.</i>
  INSERT INTO ft(ft, rowid, a, b, c) VALUES('delete', 14, $a, $b, $c);
</codeblock>

<p> If the values "inserted" into the text columns as part of a 'delete'
command are not the same as those currently stored within the table, the
results may be unpredictable.

<p> The reason for this is easy to understand: When a document is inserted
into the FTS5 table, an entry is added to the full-text index to record the
position of each token within the new document. When a document is removed,
the original data is required in order to determine the set of entries that
need to be removed from the full-text index. So if the data supplied to FTS5
when a row is deleted using this command is different from that used to
determine the set of token instances when it was inserted, some full-text 
index entries may not be correctly deleted, or FTS5 may try to remove index 
entries that do not exist. This can leave the full-text index in an
unpredictable state, making future query results unreliable.

<h2 tags="FTS5 delete-all command">The 'delete-all' Command</h2>

<p> This command is only available with [FTS5 external content tables |
external content] and [FTS5 contentless tables | contentless] tables. It
deletes all entries from the full-text index.

<codeblock>
  INSERT INTO ft(ft) VALUES('delete-all');
</codeblock>

<h2>The 'integrity-check' Command</h2>

<p> This command is used to verify that the full-text index is consistent 
with the contents of the FTS5 table or [FTS5 external content tables | content 
table]. It is not available with [FTS5 contentless tables | contentless tables].

<p>The integrity-check command is invoked by inserting the text value
'integrity-check' into the special column with the same name as the FTS5
table. For example:

<codeblock>
  INSERT INTO ft(ft) VALUES('integrity-check');
</codeblock>

<p>If the full-text index is consistent with the contents of the table, the
INSERT used to invoke the integrity-check command succeeds. Or, if any
discrepancy is found, it fails with an [SQLITE_CORRUPT_VTAB] error.

<h2 tags="FTS5 merge command">The 'merge' Command</h2>

<codeblock>
  INSERT INTO ft(ft, rank) VALUES('merge', 500);
</codeblock>

<p> This command merges b-tree structures together until roughly N pages
of merged data have been written to the database, where N is the absolute
value of the parameter specified as part of the 'merge' command. The size of
each page is as configured by the [FTS5 pgsz option].

<p> If the parameter is a positive value, B-tree structures are only eligible
for merging if one of the following is true:

<ul>
  <li> There are U or more such b-trees on a
       single level (see the documentation for the [FTS5 automerge option]
       for an explanation of b-tree levels), where U is the value assigned
       to the [FTS5 usermerge option] option.
  <li> A merge has already been started (perhaps by a 'merge' command that
       specified a negative parameter).
</ul>

<p> It is possible to tell whether or not the 'merge' command found any 
b-trees to merge together by checking the value returned by the
[sqlite3_total_changes()] API before and after the command is executed. If
the difference between the two values is 2 or greater, then work was performed.
If the difference is less than 2, then the 'merge' command was a no-op. In this
case there is no reason to execute the same 'merge' command again, at least
until after the FTS table is next updated.

<p> If the parameter is negative, and there are B-tree structures on more than
one level within the FTS index, all B-tree structures are assigned to the same
level before the merge operation is commenced. Additionally, if the parameter
is negative, the value of the usermerge configuration option is not 
respected - as few as two b-trees from the same level may be merged together.

<p> The above means that executing the 'merge' command with a negative
parameter until the before and after difference in the return value of
[sqlite3_total_changes()] is less than two optimizes the FTS index in the
same way as the [FTS5 optimize command]. However, if a new b-tree is added
to the FTS index while this process is ongoing, FTS5 will move the new 
b-tree to the same level as the existing b-trees and restart the merge. To
avoid this, only the first call to 'merge' should specify a negative parameter.
Each subsequent call to 'merge' should specify a positive value so that the
merge started by the first call is run to completion even if new b-trees are
added to the FTS index.

<h2 tags="FTS5 optimize command">The 'optimize' Command</h2>

<p>This command merges all individual b-trees that currently make up the
full-text index into a single large b-tree structure. This ensures that the
full-text index consumes the minimum space within the database and is in the
fastest form to query.

<p>Refer to the documentation for the [FTS5 automerge option] for more details
regarding the relationship between the full-text index and its component
b-trees.

<codeblock>
  INSERT INTO ft(ft) VALUES('optimize');
</codeblock>

<p>Because it reorganizes the entire FTS index, the optimize command can 
take a long time to run. The [FTS5 merge command] can be used to divide
the work of optimizing the FTS index into multiple steps. To do this:

<ul>
  <li> Invoke the 'merge' command once with the parameter set to -N, then
  <li> Invoke the 'merge' command zero or more times with the parameter set to N.
</ul>

<p>where N is the number of pages of data to merge within each invocation of
the merge command. The application should stop invoking merge when the
difference in the value returned by the sqlite3_total_changes() function before
and after the merge command drops to below two. The merge commands may be
issued as part of the same or separate transactions, and by the same or
different database clients. Refer to the documentation for the 
[FTS5 merge command | merge command] for further details.

<h2 tags="FTS5 pgsz option">The 'pgsz' Configuration Option</h2>

<p> This command is used to set the persistent "pgsz" option.

<p> The full-text index maintained by FTS5 is stored as a series of fixed-size
blobs in a database table. It is not strictly necessary for all blobs that make
up a full-text index to be the same size. The pgsz option determines the size
of all blobs created by subsequent index writers. The default value is 1000.

<codeblock>
  INSERT INTO ft(ft, rank) VALUES('pgsz', 4072);
</codeblock>

<h2 tags="FTS5 rank configuration option">The 'rank' Configuration Option</h2>

<p> This command is used to set the persistent "rank" option.

<p> The rank option is used to change the default auxiliary function mapping
for the rank column. The option should be set to a text value in the same
format as described for [auxiliary function mapping | "rank MATCH ?"] terms 
above. For example:

<codeblock>
  INSERT INTO ft(ft, rank) VALUES('rank', 'bm25(10.0, 5.0)');
</codeblock>

<h2 tags="FTS5 rebuild command">The 'rebuild' Command</h2>

<p> This command first deletes the entire full-text index, then rebuilds it
based on the contents of the table or [FTS5 external content tables | content
table].  It is not available with [FTS5 contentless tables | contentless
tables].

<codeblock>
  INSERT INTO ft(ft) VALUES('rebuild');
</codeblock>

<h2 tags="FTS5 usermerge option">The 'usermerge' Configuration Option</h2>

<p> This command is used to set the persistent "usermerge" option.

<p> The usermerge option is similar to the automerge and crisismerge options.
It is the minimum number of b-tree segments that will be merged together by
a 'merge' command with a positive parameter. For example:

<codeblock>
  INSERT INTO ft(ft, rank) VALUES('usermerge', 4);
</codeblock>

<p> The default value of the usermerge option is 4. The minimum allowed value
is 2, and the maximum 16.

<h1 tags="Extending FTS5">Extending FTS5</h1>

<p>FTS5 features APIs allowing it to be extended by:

<ul>
  <li> Adding new auxiliary functions implemented in C, and
  <li> Adding new tokenizers, also implemented in C.
</ul>

<p> The built-in tokenizers and auxiliary functions described in this
document are all implemented using the publicly available API described
below.

<p> Before a new auxiliary function or tokenizer implementation may be 
registered with FTS5, an application must obtain a pointer to the "fts5_api"
structure. There is one fts5_api structure for each database connection with
which the FTS5 extension is registered. To obtain the pointer, the application
invokes the SQL user-defined function fts5() with a single argument.  That
argument must be set to a pointer to a pointer to an fts5_api object
using the [sqlite3_bind_pointer()] interface.
The following example code demonstrates the technique:

<codeblock>
  <i>/*
  ** Return a pointer to the fts5_api pointer for database connection db.
  ** If an error occurs, return NULL and leave an error in the database 
  ** handle (accessible using sqlite3_errcode()/errmsg()).
  */</i>
  fts5_api *fts5_api_from_db(sqlite3 *db){
    fts5_api *pRet = 0;
    sqlite3_stmt *pStmt = 0;

    if( SQLITE_OK==sqlite3_prepare(db, "SELECT fts5(?1)", -1, &pStmt, 0) ){
      sqlite3_bind_pointer(pStmt, (void*)&pRet, "fts5_api_ptr", NULL);
      sqlite3_step(pStmt);
    }
    sqlite3_finalize(pStmt);
    return pRet;
  }
</codeblock>

<p><b>Backwards Compatibility Warning:</b>
Prior to SQLite version 3.20.0 ([dateof:3.20.0]), the fts5() worked slightly
differently.  Older applications that extend FTS5 must be revised to use 
the new technique shown above.

<p> The fts5_api structure is defined as follows. It exposes three methods, 
one each for registering new auxiliary functions and tokenizers, and one for
retrieving existing tokenizer. The latter is intended to facilitate the
implementation of "tokenizer wrappers" similar to the built-in
porter tokenizer.

<codeblock>
<tclscript>
  set res ""
  set ::extract_api_docs_mode fts5_api
  catch { set res [source [file join $::SRC ext/fts5/extract_api_docs.tcl]] }
  set res
</tclscript>
</codeblock>

<p> To invoke a method of the fts5_api object, the fts5_api pointer itself
should be passed as the methods first argument followed by the other, method
specific, arguments. For example:

<codeblock>
    rc = pFts5Api->xCreateTokenizer(pFts5Api, ... other args ...);
</codeblock>

<p> The fts5_api structure methods are described individually in the following
sections.

<h2 tags="custom tokenizers">Custom Tokenizers</h2>

<p> To create a custom tokenizer, an application must implement three
functions: a tokenizer constructor (xCreate), a destructor (xDelete) and a
function to do the actual tokenization (xTokenize). The type of each
function is as for the member variables of the fts5_tokenizer struct:

<codeblock>
<tclscript>
  set res ""
  set ::extract_api_docs_mode fts5_tokenizer
  catch { set res [source [file join $::SRC ext/fts5/extract_api_docs.tcl]] }
  set res
</tclscript>
</codeblock>

<p> The implementation is registered with the FTS5 module by calling the
xCreateTokenizer() method of the fts5_api object. If there is already a
tokenizer with the same name, it is replaced.  If a non-NULL xDestroy parameter
is passed to xCreateTokenizer(), it is invoked with a copy of the pContext
pointer passed as the only argument when the database handle is closed or when
the tokenizer is replaced.

<p> If successful, xCreateTokenizer() returns SQLITE_OK. Otherwise, it
returns an SQLite error code. In this case the xDestroy function is <b>not</b> 
invoked.

<p> When an FTS5 table uses the custom tokenizer, the FTS5 core calls xCreate()
once to create a tokenizer, then xTokenize() zero or more times to tokenize
strings, then xDelete() to free any resources allocated by xCreate(). More
specifically:

<tclscript>
  set res ""
  set ::extract_api_docs_mode tokenizer_api
  catch { set res [source [file join $::SRC ext/fts5/extract_api_docs.tcl]] }
  set res
</tclscript>

<h2 tags="FTS5 custom auxiliary functions">Custom Auxiliary Functions</h2>

<p> Implementing a custom auxiliary function is similar to implementing a
[application-defined SQL function | scalar SQL function]. The implementation
should be a C function of type fts5_extension_function, defined as follows:

<codeblock>
<tclscript>
  set res ""
  set ::extract_api_docs_mode fts5_extension
  catch { set res [source [file join $::SRC ext/fts5/extract_api_docs.tcl]] }
  set res
</tclscript>
</codeblock>

<p> The implementation is registered with the FTS5 module by calling the
xCreateFunction() method of the fts5_api object. If there is already an
auxiliary function with the same name, it is replaced by the new function.
If a non-NULL xDestroy parameter is passed to xCreateFunction(), it is invoked
with a copy of the pContext pointer passed as the only argument when the
database handle is closed or when the registered auxiliary function is
replaced.

<p> If successful, xCreateFunction() returns SQLITE_OK. Otherwise, it
returns an SQLite error code. In this case the xDestroy function is <b>not</b> 
invoked.

<p> The final three arguments passed to the auxiliary function callback are
similar to the three arguments passed to the implementation of a scalar SQL
function. All arguments except the first passed to the auxiliary function are
available to the implementation in the apVal&#91;&#93; array. The
implementation should return a result or error via the content handle pCtx.

<p> The first argument passed to an auxiliary function callback is a pointer
to a structure containing methods that may be invoked in order to obtain
information regarding the current query or row. The second argument is an
opaque handle that should be passed as the first argument to any such method 
invocation. For example, the following auxiliary function definition returns
the total number of tokens in all columns of the current row:

<codeblock>
<i>/*
** Implementation of an auxiliary function that returns the number
** of tokens in the current row (including all columns).
*/</i>
static void column_size_imp(
  const Fts5ExtensionApi *pApi,
  Fts5Context *pFts,
  sqlite3_context *pCtx,
  int nVal,
  sqlite3_value **apVal
){
  int rc;
  int nToken;
  rc = pApi->xColumnSize(pFts, -1, &nToken);
  if( rc==SQLITE_OK ){
    sqlite3_result_int(pCtx, nToken);
  }else{
    sqlite3_result_error_code(pCtx, rc);
  }
}
</codeblock>

<p>The following section describes the API offered to auxiliary function
implementations in detail. Further examples may be found in the "fts5_aux.c"
file of the source code.

<h3 tags="custom auxiliary functions">
  Custom Auxiliary Functions API Reference
</h3>

<codeblock>
<tclscript>
  set res ""
  set ::extract_api_docs_mode Fts5ExtensionApi
  catch { set res [source [file join $::SRC ext/fts5/extract_api_docs.tcl]] }
  set res
</tclscript>
</codeblock>

<tclscript>
  set res ""
  unset -nocomplain ::extract_api_docs_mode 
  catch { set res [source [file join $::SRC ext/fts5/extract_api_docs.tcl]] }
  set res
</tclscript>

<h1 tags="fts5vocab">The fts5vocab Virtual Table Module</h1>

<p> The fts5vocab virtual table module allows users to extract information from
an FTS5 full-text index directly. The fts5vocab module is a part of FTS5 - it 
is available whenever FTS5 is.

<p> Each fts5vocab table is associated with a single FTS5 table. An fts5vocab
table is usually created by specifying two arguments in place of column names
in the CREATE VIRTUAL TABLE statement - the name of the associated FTS5 table
and the type of fts5vocab table. Currently there are three types of fts5vocab
table; "row", "col" and "instance". Unless the fts5vocab table is created
within the "temp" database, it must be part of the same database as the
associated FTS5 table.

<codeblock>
  <i>-- Create an fts5vocab "row" table to query the full-text index belonging
  -- to FTS5 table "ft1".</i>
  CREATE VIRTUAL TABLE ft1_v USING fts5vocab('ft1', 'row');

  <i>-- Create an fts5vocab "col" table to query the full-text index belonging
  -- to FTS5 table "ft2".</i>
  CREATE VIRTUAL TABLE ft2_v USING fts5vocab(ft2, col);

  <i>-- Create an fts5vocab "instance" table to query the full-text index
  -- belonging to FTS5 table "ft3".</i>
  CREATE VIRTUAL TABLE ft3_v USING fts5vocab(ft3, instance);
</codeblock>

<p> If an fts5vocab table is created in the temp database, it may be associated
with an FTS5 table in any attached database. In order to attach the fts5vocab
table to an FTS5 table located in a database other than "temp", the name of the
database is inserted before the FTS5 table name in the CREATE VIRTUAL TABLE 
arguments. For example:

<codeblock>
  <i>-- Create an fts5vocab "row" table to query the full-text index belonging
  -- to FTS5 table "ft1" in database "main".</i>
  CREATE VIRTUAL TABLE temp.ft1_v USING fts5vocab(main, 'ft1', 'row');

  <i>-- Create an fts5vocab "col" table to query the full-text index belonging
  -- to FTS5 table "ft2" in attached database "aux".</i>
  CREATE VIRTUAL TABLE temp.ft2_v USING fts5vocab('aux', ft2, col);

  <i>-- Create an fts5vocab "instance" table to query the full-text index 
  -- belonging to FTS5 table "ft3" in attached database "other".</i>
  CREATE VIRTUAL TABLE temp.ft2_v USING fts5vocab('aux', ft3, 'instance');
</codeblock>

<p> Specifying three arguments when creating an fts5vocab table in any database
other than "temp" results in an error.

<p> An fts5vocab table of type "row" contains one row for each distinct term
in the associated FTS5 table. The table columns are as follows:

<table striped=1>
  <tr><th>Column<th>Contents
  <tr><td>term<td> The term, as stored in the FTS5 index.
  <tr><td>doc<td>  The number of rows that contain at least one instance of the term.
  <tr><td>cnt<td>  The total number of instances of the term in the entire FTS5 table.
</table>

<p> An fts5vocab table of type "col" contains one row for each distinct term/column
combination in the associated FTS5 table. Table columns are as follows:

<table striped=1>
  <tr><th>Column<th>Contents
  <tr><td>term<td> The term, as stored in the FTS5 index.
  <tr><td>col<td>  The name of the FTS5 table column that contains the term.
  <tr><td>doc<td>  The number of rows in the FTS5 table for which column $col
                   contains at least one instance of the term.
  <tr><td>cnt<td>  The total number of instances of the term that appear in
                   column $col of the FTS5 table (considering all rows). 
</table>

<p> An fts5vocab table of type "instance" contains one row for each term
instance stored in the associated FTS index. Assuming the FTS5 table is
created with the 'detail' option set to 'full', table columns are as follows:

<table striped=1>
  <tr><th>Column<th>Contents
  <tr><td>term<td>   The term, as stored in the FTS5 index.
  <tr><td>doc<td>    The rowid of the document that contains the term instance.
  <tr><td>col<td>    The name of the column that contains the term instance.
  <tr><td>offset<td> The index of the term instance within its column. Terms 
                     are numbered in order of occurrence starting from 0.
</table>

<p> If the FTS5 table is created with the 'detail' option set to 'col', then
the <i>offset</i> column of an instance virtual table always contains NULL.
In this case there is one row in the table for each unique term/doc/col 
combination. Or, if the FTS5 table is created with 'detail' set to 'none',
then both <i>offset</i> and <i>col</i> always contain NULL values. For
detail=none FTS5 tables, there is one row in the fts5vocab table for each
unique term/doc combination.

<p>Example:

<codeblock>
  <i>-- Assuming a database created using:</i>
  CREATE VIRTUAL TABLE ft1 USING fts5(c1, c2);
  INSERT INTO ft1 VALUES('apple banana cherry', 'banana banana cherry');
  INSERT INTO ft1 VALUES('cherry cherry cherry', 'date date date');

  <i>-- Then querying the following fts5vocab table (type "col") returns:
  --
  --    apple  | c1 | 1 | 1
  --    banana | c1 | 1 | 1
  --    banana | c2 | 1 | 2
  --    cherry | c1 | 2 | 4
  --    cherry | c2 | 1 | 1
  --    date   | c3 | 1 | 3
  --</i>
  CREATE VIRTUAL TABLE ft1_v_col USING fts5vocab(ft1, col);

  <i>-- Querying an fts5vocab table of type "row" returns:
  --
  --    apple  | 1 | 1
  --    banana | 1 | 3
  --    cherry | 2 | 5
  --    date   | 1 | 3
  --</i>
  CREATE VIRTUAL TABLE ft1_v_row USING fts5vocab(ft1, row);

  <i>-- And, for type "instance"
  INSERT INTO ft1 VALUES('apple banana cherry', 'banana banana cherry');
  INSERT INTO ft1 VALUES('cherry cherry cherry', 'date date date');
  --
  --    apple  | 1 | c1 | 0
  --    banana | 1 | c1 | 1
  --    banana | 1 | c2 | 0
  --    banana | 1 | c2 | 1
  --    cherry | 1 | c1 | 2
  --    cherry | 1 | c2 | 2
  --    cherry | 2 | c1 | 0
  --    cherry | 2 | c1 | 1
  --    cherry | 2 | c1 | 2
  --    date   | 2 | c2 | 0
  --    date   | 2 | c2 | 1
  --    date   | 2 | c2 | 2
  --</i>
  CREATE VIRTUAL TABLE ft1_v_instance USING fts5vocab(ft1, instance);
</codeblock>


<h1 id=appendix_a nonumber tags="comparison with fts4">
  Appendix A: Comparison with FTS3/4
</h1>

<p> Also available is the similar but more mature [fts3 | FTS3/4] module. 
FTS5 is a new version of FTS4 that includes various fixes and solutions for 
problems that could not be fixed in FTS4 without sacrificing backwards 
compatibility. Some of these problems are 
[fts5 technical differences | described below].

<h2 nonumber> Application Porting Guide </h2>

<p> In order to use FTS5 instead of FTS3 or FTS4, applications usually require
minimal modifications. Most of these fall into three categories - changes
required to the CREATE VIRTUAL TABLE statement used to create the FTS table,
changes required to SELECT queries used to execute queries against the table,
and changes required to applications that use [FTS auxiliary functions].

<h3 nonumber> Changes to CREATE VIRTUAL TABLE statements </h3>

<ol>
<li> <p>The module name must be changed from "fts3" or "fts4" to "fts5".

<li> <p>All type information or constraint specifications must be removed from
     column definitions. FTS3/4 ignores everything following the column name in
     a column definition, FTS5 attempts to parse it (and will report an error
     if it fails to).

<li> <p>The "matchinfo=fts3" option is not available. The 
     [FTS5 columnsize option | "columnsize=0"] option is equivalent.

<li> <p>The notindexed= option is not available. Adding [unindexed | UNINDEXED]
     to the column definition is equivalent.

<li> <p>The ICU tokenizer is not available.

<li> <p>The compress=, uncompress= and languageid= options are not available.
     There is as of yet no equivalent for their functionality.
</ol>

<codeblock>
  <i> -- FTS3/4 statement </i>
  CREATE VIRTUAL TABLE t1 USING fts4(
    linkid INTEGER,
    header CHAR(20),
    text VARCHAR,
    notindexed=linkid,
    matchinfo=fts3,
    tokenizer=unicode61
  );

  <i> -- FTS5 equivalent (note - the "tokenizer=unicode61" option is not</i>
  <i> -- required as this is the default for FTS5 anyway)</i>
  CREATE VIRTUAL TABLE t1 USING fts5(
    linkid UNINDEXED,
    header,
    text,
    columnsize=0
  );
</codeblock>

<h3 nonumber> Changes to SELECT statements </h3>

<ol>
  <li> <p>The "docid" alias does not exist. Applications must use "rowid"
          instead.

  <li> <p>The behaviour of queries when a column-filter is specified both as
          part of the FTS query and by using a column as the LHS of a MATCH
          operator is slightly different. For a table with columns "a" and "b"
          and a query similar to:
<codeblock>
   ... a MATCH 'b: string'
</codeblock>
       <p>FTS3/4 searches for matches in column "b". However, FTS5 always
          returns zero rows, as results are first filtered for column "b", then
          for column "a", leaving no results. In other words, in FTS3/4 the
          inner filter overrides the outer, in FTS5 both filters are applied.

  <li> <p>The FTS query syntax (right hand side of the MATCH operator) has
          changed in some ways. The FTS5 syntax is quite close to the FTS4
          "enhanced syntax". The main difference is that FTS5 is fussier 
          about unrecognized punctuation characters and similar within query
          strings. Most queries that work with FTS3/4 should also work with
          FTS5, and those that do not should return parse errors.
</ol>

<h3 nonumber> Auxiliary Function Changes </h3>

<p> FTS5 has no matchinfo() or offsets() function, and the snippet() function
is not as fully-featured as in FTS3/4. However, since FTS5 does provide 
an API allowing applications to create [custom auxiliary functions], any
required functionality may be implemented within the application code.

<p> The set of built-in auxiliary functions provided by FTS5 may be 
improved upon in the future.

<h3 nonumber> Other Issues</h3>

<ol>
  <li><p> The functionality provided by the fts4aux module is now provided
          by [fts5vocab]. The schema of these two tables is slightly different.

  <li><p> The FTS3/4 "merge=X,Y" command has been replaced by the 
          [FTS5 merge command].

  <li><p> The FTS3/4 "automerge=X" command has been replaced by the 
          [FTS5 automerge option].
</ol>

<h2 nonumber tags="fts5 technical differences"> 
  Summary of Technical Differences 
</h2>

<p>FTS5 is similar to FTS3/4 in that the primary task of each is to maintain
an index mapping from each unique token to a list of instances of that token 
within a set of documents, where each instance is identified by the document 
in which it appears and its position within that document. For example:

<codeblock>
  <i>-- Given the following SQL:</i>
  CREATE VIRTUAL TABLE ft USING fts5(a, b);
  INSERT INTO ft(rowid, a, b) VALUES(1, 'X Y', 'Y Z');
  INSERT INTO ft(rowid, a, b) VALUES(2, 'A Z', 'Y Y');

  <i>-- The FTS5 module creates the following mapping on disk:</i>
  A --&gt; (2, 0, 0)
  X --&gt; (1, 0, 0)
  Y --&gt; (1, 0, 1) (1, 1, 0) (2, 1, 0) (2, 1, 1)
  Z --&gt; (1, 1, 1) (2, 0, 1)
</codeblock>

<p>In the example above, each triple identifies the location of a token
instance by rowid, column number (columns are numbered sequentially
starting at 0 from left to right) and position within the column value (the
first token in a column value is 0, the second is 1, and so on). Using this
index, FTS5 is able to provide timely answers to queries such as "the set
of all documents that contain the token 'A'", or "the set of all documents
that contain the sequence 'Y Z'". The list of instances associated with a
single token is called an "instance-list".

<p>The principle difference between FTS3/4 and FTS5 is that in FTS3/4,
each instance-list is stored as a single large database record, whereas
in FTS5 large instance-lists are divided between multiple database records.
This has the following implications for dealing with large databases that
contain large lists:

<ul>
  <li> <p>FTS5 is able to load instance-lists into memory incrementally in
       order to reduce memory usage and peak allocation size. FTS3/4 very
       often loads entire instance-lists into memory.

  <li> <p>When processing queries that feature more than one token, FTS5 is
       sometimes able to determine that the query can be answered by
       inspecting a subset of a large instance-list. FTS3/4 almost always
       has to traverse entire instance-lists.

  <li> If an instance-list grows so large that it exceeds
       the [SQLITE_MAX_LENGTH] limit, FTS3/4 is unable to handle it. FTS5
       does not have this problem. 
</ul>

<p>For these reasons, many complex queries may use less memory and run faster 
using FTS5.

<p>Some other ways in which FTS5 differs from FTS3/4 are:

<ul>
  <li> <p>FTS5 supports "ORDER BY rank" for returning results in order of
       decreasing relevancy.

  <li> <p>FTS5 features an API allowing users to create custom auxiliary 
       functions for advanced ranking and text processing applications. The
       special "rank" column may be mapped to a custom auxiliary function
       so that adding "ORDER BY rank" to a query works as expected.

  <li> <p>FTS5 recognizes unicode separator characters and case equivalence by
       default. This is also possible using FTS3/4, but must be explicitly
       enabled.

  <li> <p>The query syntax has been revised where necessary to remove
       ambiguities and to make it possible to escape special characters
       in query terms.

  <li> <p>By default, FTS3/4 occasionally merges together two or more of the
       b-trees that make up its full-text index within an INSERT, UPDATE or
       DELETE statement executed by the user. This means that any operation
       on an FTS3/4 table may turn out to be surprisingly slow, as FTS3/4 
       may unpredictably choose to merge together two or more large b-trees
       within it. FTS5 uses incremental merging by default, which limits
       the amount of processing that may take place within any given 
       INSERT, UPDATE or DELETE operation.
</ul>

<h1 id=appendix_b nonumber tags="fts5 shadow tables">
  Appendix B: Shadow tables created by FTS5
</h1>

<p>
When an FTS5 virtual table is created in a database, between 3 and 5 real
tables are created in the database. These are known as "[shadow tables]", and are
used by the virtual table module to store persistent data. They should not
be accessed directly by the user. Many other virtual table modules, including
[FTS3] and [rtree], also create and use shadow tables.

<p>FTS5 creates the following shadow tables. In each case the actual table name
is based on the name of the FTS5 virtual table (in the following table, replace
&lt;name&gt; with the name of the virtual table to find the actual shadow
table name).

<table striped=1>
<tr><th>Table&nbsp;Name<th>Contents
<tr><td>&lt;name&gt;_data<td> This table contains most of the full-text index data.
<tr><td>&lt;name&gt;_idx<td> This table contains the remainder of the full-text
index data. It is almost always much smaller than the &lt;name&gt;_data table.
<tr><td>&lt;name&gt;_config<td> Contains the values of persistent
configuration parameters.
<tr><td>&lt;name&gt;_content<td> Contains the actual data inserted into the 
FTS5 table. This shadow table is not present for 
[FTS5 contentless tables | contentless] or 
[FTS5 external content tables|external content] FTS5 tables.
<tr><td>&lt;name&gt;_docsize<td> Contains the size of each column of each
row in the virtual table in tokens. This shadow table is not present if 
the [FTS5 columnsize option|"columnsize" option] is set to 0.
</table>