Blame | Last modification | View Log | Download
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN"><html><head><title>Tips on Writing Portable SQL for Multiple Databases for PHP</title></head><body bgcolor=white><table width=100% border=0><tr><td><h2>Tips on Writing Portable SQL </h2></td><td><div align=right><img src="cute_icons_for_site/adodb.gif"></div></td></tr></table><p>Updated 18 Sep 2003. Added Portable Native SQL section.<p>If you are writing an application that is used in multiple environments andoperating systems, you need to plan to support multiple databases. This articleis based on my experiences with multiple database systems, stretching from 4thDimension in my Mac days, to the databases I currently use, which are: Oracle,FoxPro, Access, MS SQL Server and MySQL. Although most of the advice here appliesto using SQL with Perl, Python and other programming languages, I will focus on PHP and howthe <a href="http://adodb.sourceforge.net/">ADOdb</a> database abstraction libraryoffers some solutions.<p></p><p>Most database vendors practice product lock-in. The best or fastest way todo things is often implemented using proprietary extensions to SQL. This makesit extremely hard to write portable SQL code that performs well under all conditions.When the first ANSI committee got together in 1984 to standardize SQL, the databasevendors had such different implementations that they could only agree on thecore functionality of SQL. Many important application specific requirementswere not standardized, and after so many years since the ANSI effort began,it looks as if much useful database functionality will never be standardized.Even though ANSI-92 SQL has codified much more, we still have to implement portabilityat the application level.</p><h3><b>Selects</b></h3><p>The SELECT statement has been standardized to a great degree. Nearly everydatabase supports the following:</p><p>SELECT [cols] FROM [tables]<br> [WHERE conditions]<br> [GROUP BY cols]<br> [HAVING conditions] <br> [ORDER BY cols]</p><p>But so many useful techniques can only be implemented by using proprietaryextensions. For example, when writing SQL to retrieve the first 10 rows forpaging, you could write...</p><table width="80%" border="1" cellspacing="0" cellpadding="0" align="center"><tr><td><b>Database</b></td><td><b>SQL Syntax</b></td></tr><tr><td>DB2</td><td>select * from table fetch first 10 rows only</td></tr><tr><td>Informix</td><td>select first 10 * from table</td></tr><tr><td>Microsoft SQL Server and Access</td><td>select top 10 * from table</td></tr><tr><td>MySQL and PostgreSQL</td><td>select * from table limit 10</td></tr><tr><td>Oracle 8i</td><td>select * from (select * from table) where rownum <= 10</td></tr></table><p>This feature of getting a subset of data is so useful that in the PHP classlibrary ADOdb, we have a SelectLimit( ) function that allows you to hide theimplementation details within a function that will rewrite your SQL for you:</p><pre>$connection->SelectLimit('select * from table', 10);</pre><p><b>Selects: Fetch Modes</b></p><p>PHP allows you to retrieve database records as arrays. You can choose to havethe arrays indexed by field name or number. However different low-level PHPdatabase drivers are inconsistent in their indexing efforts. ADOdb allows youto determine your prefered mode. You set this by setting the variable $ADODB_FETCH_MODEto either of the constants ADODB_FETCH_NUM (for numeric indexes) or ADODB_FETCH_ASSOC(using field names as an associative index).</p><p>The default behaviour of ADOdb varies depending on the database you are using.For consistency, set the fetch mode to either ADODB_FETCH_NUM (for speed) orADODB_FETCH_ASSOC (for convenience) at the beginning of your code. </p><p><b>Selects: Counting Records</b></p><p>Another problem with SELECTs is that some databases do not return the numberof rows retrieved from a select statement. This is because the highest performancedatabases will return records to you even before the last record has been found.</p><p>In ADOdb, RecordCount( ) returns the number of rows returned, or will emulateit by buffering the rows and returning the count after all rows have been returned.This can be disabled for performance reasons when retrieving large recordsetsby setting the global variable $ADODB_COUNTRECS = false. This variable is checkedevery time a query is executed, so you can selectively choose which recordsetsto count.</p><p>If you prefer to set $ADODB_COUNTRECS = false, ADOdb still has the PO_RecordCount() function. This will return the number of rows, or if it is not found, it willreturn an estimate using SELECT COUNT(*):</p><pre>$rs = $db->Execute("select * from table where state=$state");$numrows = $rs->PO_RecordCount('table', "state=$state");</pre><p><b>Selects: Locking</b> </p><p>SELECT statements are commonly used to implement row-level locking of tables.Other databases such as Oracle, Interbase, PostgreSQL and MySQL with InnoDBdo not require row-level locking because they use versioning to display dataconsistent with a specific point in time.</p><p>Currently, I recommend encapsulating the row-level locking in a separate function,such as RowLock($table, $where):</p><pre>$connection->BeginTrans( );$connection->RowLock($table, $where); </pre><pre><font color=green># some operation</font></pre><pre>if ($ok) $connection->CommitTrans( );else $connection->RollbackTrans( );</pre><p><b>Selects: Outer Joins</b></p><p>Not all databases support outer joins. Furthermore the syntax for outer joinsdiffers dramatically between database vendors. One portable (and possibly slower)method of implementing outer joins is using UNION.</p><p>For example, an ANSI-92 left outer join between two tables t1 and t2 couldlook like:</p><pre>SELECT t1.col1, t1.col2, t2.cola <br> FROM t1 <i>LEFT JOIN</i> t2 ON t1.col = t2.col</pre><p>This can be emulated using:</p><pre>SELECT t1.col1, t1.col2, t2.cola FROM t1, t2 <br> WHERE t1.col = t2.colUNION ALLSELECT col1, col2, null FROM t1 <br> WHERE t1.col not in (select distinct col from t2)</pre><p>Since ADOdb 2.13, we provide some hints in the connection object as to legaljoin variations. This is still incomplete and sometimes depends on the databaseversion you are using, but is useful as a general guideline:</p><p><font face="Courier New, Courier, mono">$conn->leftOuter</font>: holds theoperator used for left outer joins (eg. '*='), or false if not known or notavailable.<br><font face="Courier New, Courier, mono">$conn->rightOuter</font>: holds theoperator used for right outer joins (eg '=*'), or false if not known or notavailable.<br><font face="Courier New, Courier, mono">$conn->ansiOuter</font>: booleanthat if true means that ANSI-92 style outer joins are supported, or false ifnot known.</p><h3><b>Inserts</b> </h3><p>When you create records, you need to generate unique id's for each record.There are two common techniques: (1) auto-incrementing columns and (2) sequences.</p><p>Auto-incrementing columns are supported by MySQL, Sybase and Microsoft Accessand SQL Server. However most other databases do not support this feature. Sofor portability, you have little choice but to use sequences. Sequences arespecial functions that return a unique incrementing number every time you callit, suitable to be used as database keys. In ADOdb, we use the GenID( ) function.It has takes a parameter, the sequence name. Different tables can have differentsequences. </p><pre>$id = $connection->GenID('sequence_name');<br>$connection->Execute("insert into table (id, firstname, lastname) <br> values ($id, $firstname, $lastname)");</pre><p>For databases that do not support sequences natively, ADOdb emulates sequencesby creating a table for every sequence.</p><h3><b>Binding</b></h3><p>Binding variables in an SQL statement is another tricky feature. Binding isuseful because it allows pre-compilation of SQL. When inserting multiple recordsinto a database in a loop, binding can offer a 50% (or greater) speedup. Howevermany databases such as Access and MySQL do not support binding natively andthere is some overhead in emulating binding. Furthermore, different databases(specificly Oracle!) implement binding differently. My recommendation is touse binding if your database queries are too slow, but make sure you are usinga database that supports it like Oracle. </p><p>ADOdb supports portable Prepare/Execute with:</p><pre>$stmt = $db->Prepare('select * from customers where custid=? and state=?');$rs = $db->Execute($stmt, array($id,'New York'));</pre><p>Oracle uses named bind placeholders, not "?", so to support portable binding, we have Param() that generatesthe correct placeholder (available since ADOdb 3.92):<pre><font color="#000000">$sql = <font color="#993300">'insert into table (col1,col2) values ('</font>.$DB->Param('a').<font color="#993300">','</font>.$DB->Param('b').<font color="#993300">')'</font>;<font color="#006600"># generates 'insert into table (col1,col2) values (?,?)'# or 'insert into table (col1,col2) values (:a,:b)</font>'$stmt = $DB->Prepare($sql);$stmt = $DB->Execute($stmt,array('one','two'));</font></pre><a name="native"></a><h2>Portable Native SQL</h2><p>ADOdb provides the following functions for portably generating SQL functionsas strings to be merged into your SQL statements (some are only available sinceADOdb 3.92): </p><table width="75%" border="1" align=center><tr><td width=30%><b>Function</b></td><td><b>Description</b></td></tr><tr><td>DBDate($date)</td><td>Pass in a UNIX timestamp or ISO date and it will convert it to a datestring formatted for INSERT/UPDATE</td></tr><tr><td>DBTimeStamp($date)</td><td>Pass in a UNIX timestamp or ISO date and it will convert it to a timestampstring formatted for INSERT/UPDATE</td></tr><tr><td>SQLDate($date, $fmt)</td><td>Portably generate a date formatted using $fmt mask, for use in SELECTstatements.</td></tr><tr><td>OffsetDate($date, $ndays)</td><td>Portably generate a $date offset by $ndays.</td></tr><tr><td>Concat($s1, $s2, ...)</td><td>Portably concatenate strings. Alternatively, for mssql use mssqlpo driver,which allows || operator.</td></tr><tr><td>IfNull($fld, $replaceNull)</td><td>Returns a string that is the equivalent of MySQL IFNULL or Oracle NVL.</td></tr><tr><td>Param($name)</td><td>Generates bind placeholders, using ? or named conventions as appropriate.</td></tr><tr><td>$db->sysDate</td><td>Property that holds the SQL function that returns today's date</td></tr><tr><td>$db->sysTimeStamp</td><td>Property that holds the SQL function that returns the currenttimestamp (date+time).</td></tr><tr><td>$db->concat_operator</td><td>Property that holds the concatenation operator</td></tr><tr><td>$db->length</td><td>Property that holds the name of the SQL strlen function.</td></tr><tr><td>$db->upperCase</td><td>Property that holds the name of the SQL strtoupper function.</td></tr><tr><td>$db->random</td><td>Property that holds the SQL to generate a random number between 0.00 and 1.00.</td></tr><tr><td>$db->substr</td><td>Property that holds the name of the SQL substring function.</td></tr></table><p> </p><h2>DDL and Tuning</h2>There are database design tools such as ERWin or Dezign that allow you to generate data definition language commands such as ALTER TABLE or CREATE INDEX from Entity-Relationship diagrams.<p>However if you prefer to use a PHP-based table creation scheme, adodb provides you with this feature. Here is the code to generate the SQL to create a table with:<ol><li> Auto-increment primary key 'ID', </li><li>The person's 'NAME' VARCHAR(32) NOT NULL and defaults to '', </li><li>The date and time of record creation 'CREATED', </li><li> The person's 'AGE', defaulting to 0, type NUMERIC(16). </li></ol><p>Also create a compound index consisting of 'NAME' and 'AGE':<pre>$datadict = <strong>NewDataDictionary</strong>($connection);$flds = "<font color="#660000"> ID I AUTOINCREMENT PRIMARY,NAME C(32) DEFAULT '' NOTNULL,CREATED T DEFTIMESTAMP,AGE N(16) DEFAULT 0</font>";$sql1 = $datadict-><strong>CreateTableSQL</strong>('tabname', $flds);$sql2 = $datadict-><strong>CreateIndexSQL</strong>('idx_name_age', 'tabname', 'NAME,AGE');</pre><h3>Data Types</h3><p>Stick to a few data types that are available in most databases. Char, varcharand numeric/number are supported by most databases. Most other data types (includinginteger, boolean and float) cannot be relied on being available. I recommendusing char(1) or number(1) to hold booleans. </p><p>Different databases have different ways of representing dates and timestamps/datetime.ADOdb attempts to display all dates in ISO (YYYY-MM-DD) format. ADOdb also providesDBDate( ) and DBTimeStamp( ) to convert dates to formats that are acceptableto that database. Both functions accept Unix integer timestamps and date stringsin ISO format.</p><pre>$date1 = $connection->DBDate(time( ));<br>$date2 = $connection->DBTimeStamp('2002-02-23 13:03:33');</pre><p>We also provide functions to convert database dates to Unix timestamps:</p><pre>$unixts = $recordset->UnixDate('#2002-02-30#'); <font color="green"># MS Access date => unix timestamp</font></pre><p>The maximum length of a char/varchar field is also database specific. You canonly assume that field lengths of up to 250 characters are supported. This isnormally impractical for web based forum or content management systems. Youwill need to be familiar with how databases handle large objects (LOBs). ADOdbimplements two functions, UpdateBlob( ) and UpdateClob( ) that allow you toupdate fields holding Binary Large Objects (eg. pictures) and Character LargeObjects (eg. HTML articles):</p><pre><font color=green># for oracle </font>$conn->Execute('INSERT INTO blobtable (id, blobcol) VALUES (1,empty_blob())');$conn->UpdateBlob('blobtable','blobcol',$blobvalue,'id=1');<font color=green># non-oracle databases</font>$conn->Execute('INSERT INTO blobtable (id, blobcol) VALUES (1, null)');$conn->UpdateBlob('blobtable','blobcol',$blobvalue,'id=1');</pre><p>Null handling is another area where differences can occur. This is a mine-field,because 3-value logic is tricky.<p>In general, I avoid using nulls except for dates and default all my numericand character fields to 0 or the empty string. This maintains consistency withPHP, where empty strings and zero are treated as equivalent, and avoids SQLambiguities when you use the ANY and EXISTS operators. However if your databasehas significant amounts of missing or unknown data, using nulls might be a goodidea.<p>ADOdb also supports a portable <a href=http://phplens.com/adodb/reference.functions.concat.html#ifnull>IfNull</a> function, so you can define what to displayif the field contains a null.<h3><b>Stored Procedures</b></h3><p>Stored procedures are another problem area. Some databases allow recordsetsto be returned in a stored procedure (Microsoft SQL Server and Sybase), andothers only allow output parameters to be returned. Stored procedures sometimesneed to be wrapped in special syntax. For example, Oracle requires such codeto be wrapped in an anonymous block with BEGIN and END. Also internal sql operatorsand functions such as +, ||, TRIM( ), SUBSTR( ) or INSTR( ) vary between vendors.</p><p>An example of how to call a stored procedure with 2 parameters and 1 returnvalue follows:</p><pre> switch ($db->databaseType) {case '<font color="#993300">mssql</font>':$sql = <font color="#000000"><font color="#993333">'<font color="#993300">SP_RUNSOMETHING</font>'</font></font>; break;case '<font color="#993300">oci8</font>':$sql =<font color="#993300"> </font><font color="#000000"><font color="#993300">"declare RETVAL integer;begin :RETVAL := </font><font color="#000000"><font color="#993333"><font color="#993300">SP_RUNSOMETHING</font></font></font><font color="#993300">(:myid,:group);end;";</font> break;</font>default:die('<font color="#993300">Unsupported feature</font>');}<font color="#000000"><font color="green"> # @RETVAL = SP_RUNSOMETHING @myid,@group</font>$stmt = $db->PrepareSP($sql); <br> $db->Parameter($stmt,$id,'<font color="#993300">myid</font>');$db->Parameter($stmt,$group,'<font color="#993300">group</font>');<font color="green"># true indicates output parameter<br> </font>$db->Parameter($stmt,$ret,'<font color="#993300">RETVAL</font>',true);$db->Execute($stmt); </font></pre><p>As you can see, the ADOdb API is the same for both databases. But the storedprocedure SQL syntax is quite different between databases and is not portable,so be forewarned! However sometimes you have little choice as some systems onlyallow data to be accessed via stored procedures. This is when the ultimate portabilitysolution might be the only solution: <i>treating portable SQL as a localizationexercise...</i></p><h3><b>SQL as a Localization Exercise</b></h3><p> In general to provide real portability, you will have to treat SQL codingas a localization exercise. In PHP, it has become common to define separatelanguage files for English, Russian, Korean, etc. Similarly, I would suggestyou have separate Sybase, Intebase, MySQL, etc files, and conditionally includethe SQL based on the database. For example, each MySQL SQL statement would bestored in a separate variable, in a file called 'mysql-lang.inc.php'.</p><pre>$sqlGetPassword = '<font color="#993300">select password from users where userid=%s</font>';$sqlSearchKeyword = "<font color="#993300">SELECT * FROM articles WHERE match (title,body) against (%s</font>)";</pre><p>In our main PHP file:</p><pre><font color=green># define which database to load...</font><b>$database = '<font color="#993300">mysql</font>';include_once("<font color="#993300">$database-lang.inc.php</font>");</b>$db = &NewADOConnection($database);$db->PConnect(...) or die('<font color="#993300">Failed to connect to database</font>');<font color=green># search for a keyword $word</font>$rs = $db->Execute(sprintf($sqlSearchKeyWord,$db->qstr($word)));</pre><p>Note that we quote the $word variable using the qstr( ) function. This is becauseeach database quotes strings using different conventions.</p><p><h3>Final Thoughts</h3><p>The best way to ensure that you have portable SQL is to have your data tables designed usingsound principles. Learn the theory of normalization and entity-relationship diagrams and modelyour data carefully. Understand how joins and indexes work and how they are used to tune performance.<p> Visit the following page for more references on database theory and vendors:<a href="http://php.weblogs.com/sql_tutorial">http://php.weblogs.com/sql_tutorial</a>.Also read this article on <a href=http://phplens.com/lens/php-book/optimizing-debugging-php.php>Optimizing PHP</a>.<p><font size=1>(c) 2002-2003 John Lim.</font></body></html>