Section: User Contributed Perl Documentation (3)
SQL::SplitStatement - Split any SQL code into atomic statements
# Multiple SQL statements in a single stringmy $sql_code = <<'SQL';CREATE TABLE parent(a, b, c , d );CREATE TABLE child (x, y, "w;", "z;z");/* C-style comment; */CREATE TRIGGER "check;delete;parent;" BEFORE DELETE ON parent WHEN EXISTS (SELECT 1 FROM child WHERE old.a = x AND old.b = y)BEGIN SELECT RAISE(ABORT, 'constraint failed;'); -- Inline SQL commentEND;-- Standalone SQL; comment; with semicolons;INSERT INTO parent (a, b, c, d) VALUES ('pippo;', 'pluto;', NULL, NULL);SQL
my $sql_splitter = SQL::SplitStatement->new;my @statements = $sql_splitter->split($sql_code);
# @statements now is:## (# 'CREATE TABLE parent(a, b, c , d )',# 'CREATE TABLE child (x, y, ``w;'', ``z;z'')',# 'CREATE TRIGGER ``check;delete;parent;'' BEFORE DELETE ON parent WHEN# EXISTS (SELECT 1 FROM child WHERE old.a = x AND old.b = y)# BEGIN# SELECT RAISE(ABORT, \'constraint failed;\');# END',# 'INSERT INTO parent (a, b, c, d) VALUES (\'pippo;\', \'pluto;\', NULL, NULL)'# )
This is a simple module which tries to split any SQL
code, even includingnon-standard extensions (for the details see the ``SUPPORTED
DBMSs'' sectionbelow), into the atomic statements it is composed of.
The logic used to split the SQL code is more sophisticated than a raw "split"on the ";" (semicolon) character: first, various different statement terminatortokens are recognized (see below for the list), then this module is able tocorrectly handle the presence of said tokens inside identifiers, values,comments, "BEGIN ... END" blocks (even nested), dollar-quoted strings, MySQLcustom "DELIMITER"s, procedural code etc., as (partially) exemplified in the``SYNOPSIS'' above.
Consider however that this is by no means a validating parser (technicallyspeaking, it's just a context-sensitive tokenizer). It should rather be seenas an in-progress heuristic approach, which will gradually improve as testcases will be reported. This also means that, except for the ``LIMITATIONS''detailed below, there is no known (to the author) SQL code the most currentrelease of this module can't correctly split.
The test suite bundled with the distribution (which now includes the popularSakila and Pagila sample db schemata, as detailed in the ``SHOWCASE''section below) should give you an idea of the capabilities of this module
If your atomic statements are to be fed to a DBMS, you are encouraged to useDBIx::MultiStatementDo instead, which uses this module and also (optionally)offers automatic transactions support, so that you'll have the all-or-nothingbehavior you would probably want.
- "SQL::SplitStatement->new( %options )"
- "SQL::SplitStatement->new( \%options )"
It creates and returns a new SQL::SplitStatement object. It accepts its optionseither as a hash or a hashref.
"new" takes the following Boolean options, which for documentation purposes canbe grouped in two sets: ``Formatting Options'' and ``DBMSs Specific Options''.
A Boolean option which causes, when set to a false value (which is the default),the trailing terminator token to be discarded in the returned atomic statements.When set to a true value, the terminators are kept instead.
The possible terminators (which are treated as such depending on the context)are:
- ";" (the semicolon character);
- any string defined by the MySQL "DELIMITER" command;
- an ";" followed by an "/" (forward-slash character) on its ownline;
- an ";" followed by an "." (dot character) on its own line,followed by an "/" on its own line;
- an "/" on its own line regardless of the preceding characters(only if the "slash_terminates" option, explained below, is set).
The multi-line terminators above are always treated as a single token, that isthey are discarded (or returned) as a whole (regardless of the"slash_terminates" option value).
If your statements are to be fed to a DBMS, you are advised to keep this optionto its default (false) value, since some drivers/DBMSs don't want the terminatorto be present at the end of the (single) statement.
(Note that the last, possibly empty, statement of a given SQL text, never has atrailing terminator. See below for an example.)
An alias for the the "keep_terminators" option explained above.Note that if "keep_terminators" and "keep_terminator" are both passed to"new", an exception is thrown.
A Boolean option which causes, when set to a false value (which is the default),the spaces ("\s") around the statements to be trimmed.When set to a true value, these spaces are kept instead.
When "keep_terminators" is set to false as well, the terminator is discardedfirst (regardless of the spaces around it) and the trailing spaces are trimmedthen. This ensures that if "keep_extra_spaces" is set to false, the returnedstatements will never have trailing (nor leading) spaces, regardless of the"keep_terminators" value.
A Boolean option which causes, when set to a false value (which is the default),the comments to be discarded in the returned statements. When set to a truevalue, they are kept with the statements instead.
Both SQL and multi-line C-style comments are recognized.
When kept, each comment is returned in the same string with the atomic statementit belongs to. A comment belongs to a statement if it appears, in the originalSQL code, before the end of that statement and after the terminator of theprevious statement (if it exists), as shown in this pseudo-SQL snippet:
/* This comment will be returned together with statement1 */ <statement1>; -- This will go with statement2 -- (note the semicolon which closes statement1) <statement2> -- This with statement2 as well
A Boolean option which causes, when set to a false value (which is the default),the empty statements to be discarded. When set to a true value, the emptystatements are returned instead.
A statement is considered empty when it contains no characters other than theterminator and space characters ("\s").
A statement composed solely of comments is not recognized as empty and maytherefore be returned even when "keep_empty_statements" is false. To avoidthis, it is sufficient to leave "keep_comments" to false as well.
Note instead that an empty statement is recognized as such regardless of thevalue of the options "keep_terminators" and "keep_extra_spaces".
These options are basically to be kept to their default (false) values,especially if the atomic statements are to be given to a DBMS.
They are intended mainly for cosmetic reasons, or if you want to count by howmany atomic statements, including the empty ones, your original SQL code wascomposed of.
Another situation where they are useful (in the general case necessary, really),is when you want to retain the ability to verbatim rebuild the original SQLstring from the returned statements:
my $verbatim_splitter = SQL::SplitStatement->new( keep_terminators => 1, keep_extra_spaces => 1, keep_comments => 1, keep_empty_statements => 1 ); my @verbatim_statements = $verbatim_splitter->split($sql_string); $sql_string eq join '', @verbatim_statements; # Always true, given the constructor above.
Other than this, again, you are recommended to stick with the defaults.
DBMSs Specific Options
The same syntactic structure can have different semantics across different SQLdialects, so sometimes it is necessary to help the parser to make the rightdecision. This is the function of these options.
A Boolean option which causes, when set to a true value (which is the default),a "/" (forward-slash) on its own line, even without a preceding semicolon,to be admitted as a (possible) terminator.
If set to false, a forward-slash on its own line is treated as a statementterminator only if preceded by a semicolon or by a dot and a semicolon.
If you are dealing with Oracle's SQL, you should let this option set, since aslash (alone, without a preceding semicolon) is sometimes used as a terminator,as it is permitted by SQL*Plus (on non-block statements).
With SQL dialects other than Oracle, there is the (theoretical) possibility thata slash on its own line can pass the additional checks and be considered aterminator (while it shouldn't). This chance should be really tiny (it has neverbeen observed in real world code indeed). Though negligible, by setting thisoption to false that risk can anyway be ruled out.
- "$sql_splitter->split( $sql_string )"
This is the method which actually splits the SQL code into its atomiccomponents.
It returns a list containing the atomic statements, in the same order theyappear in the original SQL code. The atomic statements are returned according tothe options explained above.
Note that, as mentioned above, an SQL string which terminates with a terminatortoken (for example a semicolon), contains a trailing empty statement: this iscorrect and it is treated accordingly (if "keep_empty_statements" is set to atrue value):
my $sql_splitter = SQL::SplitStatement->new( keep_empty_statements => 1 ); my @statements = $sql_splitter->split( 'SELECT 1;' ); print 'The SQL code contains ' . scalar(@statements) . ' statements.'; # The SQL code contains 2 statements.
- "$sql_splitter->split_with_placeholders( $sql_string )"
It works exactly as the "split" method explained above, except that it returnsalso a list of integers, each of which is the number of the placeholderscontained in the corresponding atomic statement.
More precisely, its return value is a list of two elements, the first of whichis a reference to the list of the atomic statements exactly as returned by the"split" method, while the second is a reference to the list of the number ofplaceholders as explained above.
Here is an example:
# 4 statements (valid SQLite SQL)my $sql_code = <<'SQL';CREATE TABLE state (id, name);INSERT INTO state (id, name) VALUES (?, ?);CREATE TABLE city (id, name, state_id);INSERT INTO city (id, name, state_id) VALUES (?, ?, ?)SQL
my $splitter = SQL::SplitStatement->new;
my ( $statements, $placeholders )
= $splitter->split_with_placeholders( $sql_code );
# $placeholders now is: [0, 2, 0, 3]
where the returned $placeholders list(ref) is to be read as follows: thefirst statement contains 0 placeholders, the second 2, the third 0 and thefourth 3.
The recognized placeholders are:
- question mark placeholders, represented by the "?" character;
- dollar sign numbered placeholders, represented by the"$1, $2, ..., $n" strings;
- named parameters, such as ":foo", ":bar", ":baz" etc.
- "$sql_splitter->keep_terminators( $boolean )"
Getter/setter method for the "keep_terminators" option explained above.
An alias for the "keep_terminators"
method explained above.
- "$sql_splitter->keep_extra_spaces( $boolean )"
Getter/setter method for the "keep_extra_spaces" option explained above.
- "$sql_splitter->keep_comments( $boolean )"
Getter/setter method for the "keep_comments" option explained above.
- "$sql_splitter->keep_empty_statements( $boolean )"
Getter/setter method for the "keep_empty_statements" option explained above.
- "$sql_splitter->slash_terminates( $boolean )"
Getter/setter method for the "slash_terminates" option explained above.
SQL::SplitStatement aims to cover the widest possible range of DBMSs, SQL
dialects and extensions (even proprietary), in a (nearly) fully transparent wayfor the user.
Currently it has been tested mainly on SQLite, PostgreSQL, MySQL and Oracle.
Procedural code is by far the most complex to handle.
Currently any block of code which start with "FUNCTION", "PROCEDURE","DECLARE", "CREATE" or "CALL" is correctly recognized, as well asanonymous "BEGIN ... END" blocks, dollar quoted blocks and blocksdelimited by a "DELIMITER"-defined custom terminator, therefore a wide rangeof procedural extensions should be handled correctly. However, only PL/SQL,PL/PgSQL and MySQL code has been tested so far.
If you need also other procedural languages to be recognized, please let me know(possibly with some test cases).
Bound to be plenty, given the heuristic nature of this module (and its ambitiousgoals). However, no limitations are currently known.
Please report any problematic test case.
To be split correctly, the given input must, in general, be syntactically validSQL.
For example, an unbalanced "BEGIN"
or a misspelled keyword could, undercertain circumstances, confuse the parser and make it trip over the nextstatement terminator, thus returning non-split statements.This should not be seen as a limitation though, as the original (invalid) SQL
code would have been unusable anyway (remember that this is NOT
To test the capabilities of this module, you can run it(or rather run sql-split) on the files t/data/sakila-schema.sql
included in the distribution, which contain twoquite large and complex real world
db schemata, for MySQL and PostgreSQLrespectively.
For more information:
- Sakila db: <http://dev.mysql.com/doc/sakila/en/sakila.html>
- Pagila db: <http://pgfoundry.org/projects/dbsamples>
SQL::SplitStatement depends on the following modules:
- SQL::Tokenizer 0.22 or newer
Emanuele Zeppieri, "<emazepAATTcpan.org>"
No known bugs.
Please report any bugs or feature requests to"bug-sql-SplitStatement at rt.cpan.org", or through the web interface at<http://rt.cpan.org/NoAuth/ReportBug.html?Queue=SQL-SplitStatement>.I will be notified, and then you'll automatically be notified of progresson your bug as I make changes.
You can find documentation for this module with the perldoc command:
You can also look for information at:
- RT: CPAN's request tracker
- AnnoCPAN: Annotated CPAN documentation
- CPAN Ratings
- Search CPAN
Igor Sutton for his excellent SQL::Tokenizer, which made writingthis module a joke.
LICENSE AND COPYRIGHT
Copyright 2010-2011 Emanuele Zeppieri.
This program is free software; you can redistribute it and/or modify itunder the terms of either: the GNU General Public License as publishedby the Free Software Foundation, or the Artistic License.
See http://dev.perl.org/licenses/ for more information.
- SUPPORTED DBMSs
- Procedural Extensions
- SEE ALSO
- LICENSE AND COPYRIGHT
This document was created byman2html,using the manual pages.