NAME DBD::PgLite - PostgreSQL emulation mode for SQLite SUMMARY use DBI; my $dbh = DBI->connect('dbi:PgLite:dbname=file'); # The following PostgreSQL-flavoured SQL is invalid # in SQLite directly, but works using PgLite my $sql = q[ SELECT news_id, title, cat_id, cat_name, sc_id sc_name, to_char(news_created,'FMDD.FMMM.YYYY') AS ndate FROM news NATURAL JOIN x_news_cat NATURAL JOIN cat NATURAL JOIN subcat WHERE news_active = TRUE AND news_created > NOW() - INTERVAL '7 days' ]; my $res = $dbh->selectall_arrayref($sql,{Columns=>{}}); # From v. 0.05 with full sequence function support my $get_nid = "SELECT NEXTVAL('news_news_id_seq')"; my $news_id = $dbh->selectrow_array($get_nid); DESCRIPTION The module automatically and transparently transforms a broad range of SQL statements typical of PostgreSQL into a form suitable for use in SQLite. This involves both (a) parsing and filtering of the SQL; and (b) the addition of several PostgreSQL-compatible functions to SQLite. Mainly because of datatype issues, support for many PostgreSQL features simply cannot be provided without elaborate planning and detailed metadata. Since this module is intended to be usable with any SQLite3 database, it follows that the emulation is limited in several respects. An overview of what works and what doesn't is given in the following section on PostgreSQL Compatibility. DBD::PgLite has support of a sort for stored procedures. This is described in the Extras section below. So are the few database functions defined by this module which are not in PostgreSQL. Finally, the Extras section contains a brief mention of the DBD::PgLite::MirrorPgToSQLite companion module. If you do not want SQL filtering to be turned on by default for the entire session, you can connect setting the connection attribute *FilterSQL* to a false value: my $dbh = DBI->connect("dbi:PgLite:dbname=$fn", undef, undef, {FilterSQL=>0}); To turn filtering off (or on) for a single statement, you can specify *FilterSQL* option as a statement attribute, e.g.: $dbh->do($sql, {FilterSQL=>0}, @bind); my $sth = $dbh->prepare($sql, {FilterSQL=>0}); $res = $dbh->selectall_arrayref($sql, {FilterSQL=>0}, @bind); It is possible to specify user-defined pre- and postfiltering routines, both globally (by specifying them as attributes of the database handle) and locally (by specifying them as statement attributes): $dbh = DBI->connect("dbi:PgLite:$file",undef,undef, {prefilter=>\&prefilter}); $res = $dbh->selectall_arrayref($sql, {postfilter=>\&postfilter}, @bind_values); The pre-/postfiltering subroutine receives the SQL as parameter and is expected to return the changed SQL. STATUS OF THE MODULE This module was initially developed using SQLite 3.0 and PostgreSQL 7.3, but it should be fully compatible with newer versions of both SQLite (3.1 and 3.2 have been tested) and PostgreSQL (8.1 has been tested). Support for SELECT statements and the WHERE-conditions of DELETE and UPDATE statements is rather good, though still incomplete. The module especially focuses on NATURAL JOIN differences and commonly used, built-in PostgreSQL functions. Support for inserted/updated values in INSERT and UPDATE statements could use some improvement but is useable for simple things. There is no support for differences in DDL. The SQL transformations used are not based on a formal grammar but on applying simple regular expressions. An obvious consequence of this is that they may depend excessively on the author's SQL style. YMMV. (I would however like you to contact me if you come across some SQL statements which you feel should work but that don't). The development of this module has been driven by personal needs, and so is likely to be even more one-sided than the above description suggests. POSTGRESQL COMPATIBILITY In this section, the PostgreSQL functions and operators supported by the module are enumerated. Regex operators * The regex operators "~", "~*", "!~" and "!~*" are transformed into calls to the user-defined function matches(). The regex flavour supported is Perl, not plain vanilla POSIX, so some incompatibilities may arise. * Note that for ease of parsing, whitespace before and after the operator is required for the filtering to succeed. So "col ~ 'pat'" works, but "col~'pat'" doesn't. * "SIMILAR TO" is not supported. * ILIKE is quietly changed to LIKE. LIKE in SQLite is case-insensitive for 7-bit characters. In future, ILIKE will probably be handled more elegantly, and LIKE will be redefined so as to be more like PostgreSQL. Math Functions * Added: abs, cbrt, ceil, degrees, exp, floor, ln, log (1- and 2-argument forms), mod, pi, pow, radians, sign, sqrt, trunc (1- and 2-argument forms), acos, asin, atan, atan2, cos, cot, sin, tan. * random() exists in SQLite but was redefined to conform better with PostgreSQL in terms of value range. setseed() was also added, but is not entirely compatible with PostgreSQL in the sense that setting the random seed does not engender the same sequence of pseudo-random numbers as it would in PostgreSQL. * SQLite already has a handful of mathematical functions which have been left alone, notably round() (1- and 2-argument forms). String Functions The only string functions which are present natively in SQLite are substr(), lower() and upper(). These have been left alone. Added functions are the following: * ascii, bit_length, btrim, char_length, character_length, chr, convert (1- and 2-arg), decode, encode, initcap, length, lpad, ltrim, md5, octet_length, position, pg_client_encoding (always 'SQL_ASCII'), quote_ident, quote_literal, repeat, replace, rpad, rtrim, split_part, strpos, substring(string,offset,length), substring(string from pattern), to_ascii (assumes latin-1 input), to_hex, translate, trim. Except for convert(), where another input encoding can be specified explicitly, these functions all assume that the strings are in an 8-bit character set, preferably iso-8859-1. The little-used idiom "substring(string from pattern for escape)" (where 'pattern' is not a POSIX regular expression but a SQL pattern) is not supported. Otherwise support for string functions is pretty complete. Data Type Formatting Functions The implementation of these functions is impeded by the sparse type system employed by SQLite. Workarounds are possible, however, so this area will probably be better covered in future. * to_char(timestamp, format) is mostly supported. There is support for most formatting strings (all except 'Y,YYY', 'IYYY', 'IYY', 'IY', 'I', 'J', 'TZ', and 'tz'). The FM prefix is supported for 'MM', 'DD' and 'HH*', but not otherwise. Other prefixes are not supported. * to_char(interval, format) and to_char(number, format) are not currently supported. Nor are to_date(), to_timestamp() and to_number() (yet). Date/Time Functions Again, SQLite's intrinsically bad support for dates and intervals makes this area somewhat hard to cover properly. Function support is as follows; also note the caveats below: * Supported: now, current_date, current_time, current_datetime, date_part (with timestamps, not intervals), date_trunc, extract (with timestamps, not intervals), localtime, localtimestamp, timeofday. * Not supported: age, isfinite, overlaps. Versions of SQLite 3.1 and later support some of these functions, e.g. current_date. In these versions the built-in will be overridden. The module makes no distinction between time/timestamp with and without time zone. It is assumed that times and timestamps are either all GMT or all localtime; time zone information is silently discarded. This may change later. Support for calculations with dates and intervals is still very limited. Basically, what is supported are expressions of the form "expr +/- interval 'descr'" where expr reduces to a timestamp or date value. If a transaction is started with begin_work(), the time as represented by now() and friends is "frozen" in the same way as in PostgreSQL until commit() or rollback() are called. A transaction started by simply running the SQL statement "BEGIN" does not, however, trigger this behaviour. Nor is the time automatically "unfrozen" when an error occurs during a transaction; you need to catch exceptions and call rollback() manually. Sequence Manipulation Functions * There is now full support for all explicit invocations of the sequence functions nextval(), setval(), currval() and lastval(). Sequences are emulated using the table pglite_seq. (This works even with multiple connections to the same database file, some of which are using transactions, since SQLite transactions lock the whole database file, luckily eliminating any risk of two connections getting the same value from a nextval() call). Please be aware that sequences are autogenerated if they do not exist. Be careful to specify the appropriate sequence names or you will get unexpected results. If a sequence being autogenerated ends with '_seq' and has a name which seems to match an existing table + an integer column from that table (tablename_colname_seq), it is given an initial value based on the maximum value in the column in question. There is as yet no support for CREATE SEQUENCE statements. Use the autogeneration feature to create sequences. Implicit calls to NEXTVAL() by omitting the serial column from the column list in an INSERT are caught in most cases. The main conditions that must be fulfilled for this to work are: (1) that the column in question is an integer column which is the sole primary key on the table; and (2) that the statement is a normal INSERT with a column list and a VALUES clause (and not, e.g., a statement of the form INSERT INTO x SELECT * FROM y). There is as yet no interaction with the SQLite builtin autoincrement/last_insert_rowid() functionality in connection with the sequence function support. Aggregate Functions * max(), min(), count() and sum() are already supported by SQLite and have been left alone. Note that the construct "count(distinct colname)" is not supported unless the SQLite version being used supports it (3.2.6 and later). * avg() has been added. * stddev() and variance() are not supported. A Note on Casting Casting using the construct "::datatype" is not supported in general. However, "::int", "::date" and "::bool" should work as expected. All other casts are silently discarded. A Note on Booleans This module assumes that booleans will be stored as numeric values in the SQLite database. SQLite interprets 0 as false and any non-zero numeric value as true. Accordingly, expressions such as "= TRUE" and "= 't'" are simply removed in SELECT and DELETE statements. Likewise, "expr = FALSE" is turned into "NOT expr" before being passed on to SQLite. In INSERT and DELETE statements, TRUE and FALSE (as well as 't'::bool and 'f'::bool - but not 't' and 'f' by themselves) are turned into 1 and 0. Current_user etc. The functions current_user(), session_user() and user() - with or without parentheses - all mean the same thing. They return the username of the effective uid. Other Functions The main groups of other functions (not supported by this module at all) are: * Database/user information functions: Aside from current_user/session_user/user, which were mentioned above, no functions in this group are supported. This includes current_database(), current_schema(), all functions with names starting with 'pg_' and 'has_', obj_description and col_description. See http://www.postgresql.org/docs/current/static/functions-misc.html * Array functions are not implemented - see http://www.postgresql.org/docs/current/static/functions-array.html * Binary string (BYTEA) functions are not implemented - see http://www.postgresql.org/docs/current/static/functions-binarystring .html * Geometric functions are not implemented - see http://www.postgresql.org/docs/current/static/functions-geometry.htm l * Network Address Functions are not implemented - see http://www.postgresql.org/docs/current/static/functions-net.html EXTRAS Stored Procedures If the active database file contains a table called pglite_functions, the module assumes that it will have the following structure: CREATE TABLE pglite_functions ( name TEXT, -- name of the function argnum INT, -- number of arguments (-1 means any number) type TEXT, -- can be 'sql' or 'perl' sql TEXT, -- the body of the function PRIMARY KEY (name, argnum) ); In the case of a SQL-type function, it can contain syntax supported through the module (and not directly by SQLite). The numeric arguments ($1-$9) customary in PostgreSQL are supported, so that in many cases simple functions will be directly transferrable from pg_proc in a PostgreSQL database. An instance of a SQL snippet which would work as a function body both in PostgreSQL and PgLite (e.g. with the function name 'full_price_descr'): SELECT TRIM(group_name||': '||price_description) FROM price_group NATURAL JOIN price WHERE price_id = $1 As for perl-type functions, the function body is simply the text of a subroutine. Here is a simple example of a function body for the function 'commify', which takes two arguments: the number to be formatted and the desired number of decimal places: sub { my ($num,$dp) = @_; my $format = "%.${dp}f"; $num = scalar reverse(sprintf $format, $num); my $rest = $1 if $num =~ s/^(\d+)\.//; $num =~ s/(...)/$1,/g; $num = "$rest.$num" if $rest; return scalar reverse($num); } Non-Pg Functions matches(), imatches(): These functions are used behind the scenes to implement support for the '~' regex-matching operator and its variants. They take two arguments, a string and a regular expression. matches() is case sensitive, imatches() isn't. matches_safe(), imatches_safe(): These work in the same way as matches() and imatches() except that metacharacters are escaped in the regex argument. They are therefore in many cases more suitable for user input and other untrusted sources. lower_latin1(): Depending on platform, lower() and upper() may not transform the case of non-ascii characters despite a proper locale being defined in the environment. This functions assumes that a Latin-1 locale is active and returns a lower-case version of the input given this assumption. localeorder(): DBD::SQLite does not provide access to defining SQLite collation functions. This is a workaround for a specific case where this limitation can be an issue. Given a Latin-1 encoded string, it returns a string of hex digits which can be ascii-sorted in the ordinary way. The resulting row order will be in accordance with the currently active locele - but only if the locale is Latin-1 based. The sort is case-insensitive. locale(): An information function simply returning the name of the current locale. The module sets the locale based on the environment variables $ENV{LC_COLLATE}, $ENV{LC_ALL}, $ENV{LANG}, and $ENV{LC_CTYPE}, in that order. Currently it is not possible to use different locales for character type and collation, as far as the module is concerned. DBD::PgLite::MirrorPgToSQLite The companion module, DBD::PgLite::MirrorPgToSQLite, may be of use in conjunction with this module. It can be used for easily mirroring specific tables from a PostgreSQL database, moving views and (some) functions as well if desired. CAVEATS Some functions defined by the module are not suitable for use with UTF-8 data and/or in an UTF-8 locale. (This, however, would be rather easy to change if you're willing to sacrifice proper support for 8-bit locales such as iso-8859-1). Please do not make the mistake of using this module for an important production system - too much can go wrong. But as a development tool it can be useful, and as a toy it can be fun... TODO There is a lot left undone. The next step is probably to handle non-SELECT statements better. SEE ALSO DBI, DBD::SQLite, DBD::Pg, DBD::PgLite::MirrorPgToSQLite; THANKS TO Johan Vromans, for encouraging me to improve the sequence support. AUTHOR Baldur Kristinsson (bk@mbl.is), 2006. Copyright (c) 2006 Baldur Kristinsson. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself. =================================== NAME DBD::PgLite::MirrorPgToSQLite - Mirror tables from PostgreSQL to SQLite SUMMARY use DBD::PgLite::MirrorPgToSQLite qw(pg_to_sqlite); pg_to_sqlite( sqlite_file => '/var/pg_mirror/news.sqlite', pg_dbh => $dbh, schema => 'news', tables => [ qw(news cat img /^x_news/)], views => [ 'v_newslist' ], indexes => 1, verbose => 1, snapshot => 1, ); USAGE The purpose of this module is to facilitate mirroring of tables from a PostgreSQL dataabse to a SQLite file. The module has only be tested with PostgreSQL 7.3 and SQLite 3.0-3.2. SQLite 2.x will probably not work; as for PostgreSQL, any version after 7.2 is supposed to work. If it doesn't, please let me know. As seen above, options to the pg_to_sqlite() function (which is exported on request) are passed in as a hash. These options are described below. The default values can be changed by overriding the DBD::PgLite::MirrorPgToSQLite::defaults() subroutine. Required options Obviously, the mirroring function needs either a PosgtgreSQL database connection or enough information to be able to connect to the database by itself. It also needs the name of a target SQLite file, and a list of tables to copy between the two databases. pg_dbh, pg_user, pg_pass, pg_dsn If a database handle is specified in *pg_dbh*, it takes precedence. Otherwise we try to connect using *pg_dsn*, *pg_user*, and *pg_pass* (which are assigned defaults based on the environment variables PGDATABASE, PGUSER and PGPASSWORD, if any of these is present). tables The value of the required *tables* option should be an arrayref of strings or a string containing a comma-separated list of tablenames and tablename patterns. A tablename pattern is a string or distinct string portion delimited by forward slashes. To clarify: Suppose that a database contains the tables news, img, img_group, cat, users, comments, news_read_log, x_news_cat, x_news_img, and x_img_group; and that we want to mirror news, img, cat, x_news_img and x_news_cat, leaving the other tables alone. To achieve this, you would set the *tables* option to any of the following (there are of course also other possibilities): (1) [qw(news img cat x_news_img x_news_cat)] (2) 'news, img, cat, x_news_img, x_news_cat' (3) [qw(news /img$/ /cat$/)] (4) 'news,/img$/,/cat/' The purpose of this seemingly unneccesary flexibility in how the table list is specified is to make the functionality of the module more easily accessible from the command line. Please note that the patterns between the slash delimiters are not Perl regular expressions but rather POSIX regular expressions, used to query the PostgreSQL system tables directly. sqlite_file This should specify the full path to a SQLite file. While the mirroring takes place, the incoming data is not written directly to this file, but to a file with the same name except for a '.tmp' extension. When the operation has finished, the previous file with the name specified (if any) is renamed with a '.bak' extension, and the .tmp file is renamed to the requested filename. Unless you use the *append* option, the information previously in the file will be totally replaced. Other options schema This signifies the schema from which the tables on the PostgreSQL side are to be fetched. Default: 'public'. Only one schema can be specified at a time. where A WHERE-condition appended to the SELECT-statement used to get data from the PostgreSQL tables. views A list of views, specified in the same manner as the list of tables for the *tables* option. An attempt is made to define corresponding views on the SQLite side (though this functionality is far from reliable). indexes A boolean option indicating whether to create indexes for the same columns in SQLite as in PostgreSQL. Default: false. (Normally only the primary key is created). functions A boolean indicating whether to attempt to create functions on the SQLite side corresponding to any SQL language (NOT PL/pgSQL or other procedural language) functions in the PostgreSQL database. This is for use with DBD::PgLite only, since these functions are put into the pglite_functions table. Default: false. page_limit Normally the information from the PostgreSQL tables is read into memory in one go and transferred directly to the SQLite file. This is, however, obviously not desireable for very large tables. If the PostgreSQL system tables report that the page count for the table is above the limit specified by *page_limit*, the table is instead transferred row-by-row. Default value: 5000; since each page normally is 8K, this represents about 40 MB on disk and perhaps 70-100 MB of memory usage by the Perl process. For page_limit to work, the table must have a primary key. NB! Do not set this limit lower than necessary: it is orders of magnitude slower than the default "slurp into memory" mode. append If this boolean option is true, then instead of creating a new SQLite file, the current contents of the *sqlite_file* are added to. If a table which is being mirrored existed previously in the file, it is dropped and recreated, but any tables not being copied from PostgreSQL in the current run are left alone. (This is primarily useful for mirroring some tables in toto, and others only in part, into the same file). Default: false. Incompatible with the *snapshot* option. snapshot If this is true, then the copying from PostgreSQL takes place in serialized mode (transaction isolation level serializable), which should ensure consistency of relations between tables linked by foreign key constraints. Currently, foreign keys are not created on the SQLite side, however. Default: false. Incompatible with the *append* option. cachedir The current method for getting information about table structure in PostgreSQL is somewhat slow, especially for databases with very many tables. To offset this, table definitions are cached in a temporary directory so that subsequent mirrorings of the same table will go faster. The downside is, of course, that if the table structure changes, the cache needs to be cleared manually. The cache directory can be specified using this option; the default is /tmp/sqlite_mirror_cache (with separate subdirectories for each user). verbose If this is true, a few messages will be output to stderr during the mirroring process. TODO * Support for foreign keys is missing. * The method used to read tables bigger than *page_limit* needs to be improved. * It would be nice to have a quick way of telling whether the cached table definition of a specific table is still valid. * Tests. AUTHOR Baldur Kristinsson (bk@mbl.is), 2004-2006. Copyright (c) 2006 Baldur Kristinsson. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.