I’ve recently been contracted by Geekcruises(http://www.geekcruises.com) to update their onlinebooking and administration forms. They’re using a simple CGIarchitecture, with a lot of hand-rolled SQL to talk to theirPostgreSQL database.
My first design decision was to adopt a more generic flexibleweb abstraction, creating CGI::Prototype as a result.They’ve allowed me to upload that to the CPAN, and I’ll be writingmore about that in the future. For the “model” part of theabstraction, I’ve settled in on the wildly popularClass::DBI framework. And that brings us to today’sproblem and solution.
The existing database consists of 50 tables, with typicalforeign key mappings. For example, a person_cruiserecord links a person with a cruise. InClass::DBI, setting up these foreign_key relationshipsso that they could be easily accessed looked simple enough: I’djust have to set up a bunch of statements like:
And that was fine, for the first 10 or 15 records. And then itstarted to get boring. And then tedious. And then error-prone. Istarted to think “hey, wait a minute! PostgreSQL knows theserelationships! Why am I reentering that information?” And then“Oh, someone else must’ve done this already!”
When I started looking in the CPAN for some reusable solutions,I quickly found Class::DBI::Loader. Joy! This modulesolves the problem of querying the database for all the records andtheir columns, including determining the auto increment sequences.That’s good, because that was already starting to be frustrating aswell.
But Class::DBI::Loader did nothing about theforeign keys. So, as I kept looking, I stumbled uponClass::DBI::Loader::Relationship, which started tolook promising (I could use English instead of Perl), but alas,even the example has a bug in it, and I just couldn’t get it towork like I wanted. Also, again, why am I specifying all theseforeign keys when the database itself already knows?
After staring at the DBD::Pg manpage for about ahalf hour, and trying out little snippets in the debugger, I wasfinally able to coax the database into telling me about its foreignkeys. Then I took some of the ideas fromClass::DBI::Loader::Relationship to turn those intohas_a and has_many calls (including theidea of automatically making plurals). And after a bit of hacking,I had my automatic Class::DBI generator.
However, I ran into one other snag. The database had usedPostgreSQL’s table inheritance for the payment types. A specificcredit card, money order, or check payment inherited from a genericpayment record, so that an SQL query could ask for “any payment”versus “any credit card payment” easily. Since these child recordtypes didn’t have an explicit foreign-key relationship with thelinking records, I had to add a means of faking it in my code.
The result of all of this is in [listing one below]. Lines 1 and94 define this as a package GC::DB, which I willrequire or use in my application.
Lines 4 through 90 create a lexical scope so that I can definetemporary variables during the setup that won’t apply to any of thelater code.
Lines 6 and 7 set up the database connection, user, and passwordfor this database. They can be overridden for testing purposes bysetting @GC::DB::CONNECTION before bringing in thisfile. More on that in a future column as well.
Lines 9 and 10 bring in two CPAN modules. I’m usingrequire rather than use because I don’twant any imports to pollute GC::DB class.
Line 12 turns on debug tracing mode. This was helpful when I wasdeveloping, to see what relationships were actually beinggenerated.
Lines 14 through 18 use Class::DBI::Loader tocreate the 50 Class::DBI classes reflecting the 50records of the database. The namespace parameter givesGC::DB:: prefixes to these new classes. This call alsosets up all the sequence calls so that key generationworks properly.
Lines 19 and 20 extract out the classes and tables justgenerated. I’m presuming the two lists correlate, but have noassurance of that from the docs. In fact, the docs don’t evenmention these two methods.
Line 22 gets me a database handle by callingdb_Main on the first class. Any class would have done,as they all share the same database.
Lines 26 to 29 add our current class to the @ISAlist for each of the created classes. I want to be able to definemethods that all of my created classes understand, and this was thefastest way to do that.
Lines 33 and 34 set up the %class_of hash, so thatI can go from a table name to a class name. Again, I’m presumingthat the two lists are correlated properly.
Line 36 patches up a plural that didn’t come out quite right forme. The class named GC::DB::PaidTo was getting alinking method called paidtoes, which was humorous butexpectedly hard to remember. So, I just whacked the pluralizer onthe head with an additional fact, and it’s all good.
Lines 38 to 60 define a “lexical subroutine” as a coderefstored in a lexical variable. Again, I’m trying hard not to pollutethe GC::DB namespace. This subroutine takes a tablename, a linking column in that table, and the other table to whichthe column links, and turns those into the right has_aand has_many calls.
Lines 41 and 42 use the class name hash to get the correspondingclass names for the two table names.
Lines 44 through 48 provide a debugging trace of thehas_a call that we’re about to execute in line 49.This creates our forward link (many to one).
Lines 51 and 52 determine the plural reverse link by firstextracting the “basename” of the class, then lowercasing that,then pluralizing that.
Lines 54 through 58 provide a debugging trace of thehas_many call that we’re about to execute in line 59.This creates our reverse link (one to many).
And now that we have that subroutine to do the dirty work, wejust need to figure out how to call it!
Lines 64 through 68 define a hash to map a table name into allof its child tables, which apparently we cannot determineautomatically from the DBI layer. Oh well.
Lines 70 through 89 loop through all the tables we’ve seen. Line71 traces the particular table being examined.
Line 72 uses the DBI abstraction to get the foreignkeys for a given table (indicated as the sixth positionalparameter). The return value from foreign_key_info isa statement handle that can be used with the normalDBI methods to extract the data. If there are noforeign keys in this table, $sth is undefined, so weneed to skip the rest of the loop.
Lines 73 through 87 loop through the results from getting thedetails of the foreign keys as an array of hashrefs. Each$res is a hashref for a particular foreign keymapping.
Lines 75 and 76 extract the linking column and the referencedtable from the result.
Lines 78 and 79 deal with the PostgreSQL inheritance by possiblyreplacing a single table name with all of its related tables. Mostof the time, @tables will consist of only$table, but when $table waspayment, we’ll instead have the entire list.
Lines 81 through 85 execute the $has_a_manysubroutine for each combination of source table, foreign table, andcolumn that we’ve now noted.
Line 92 is where I will be adding other things for each class.For example, if I need to add a method to theGC::DB::PersonCruise record, I’ll merely add:
to the mix, keeping all of my database model code in this onefile.
The lesson I’m reminded here is that when things seem tedious,it’s time to get the computer to do more of the work. And I’m happythat I have a language like Perl that can generate and execute codeon the fly, including late-binding the method calls to the objectsand classes created at run-time. Try that in Java. Not gonnahappen. Until next time, enjoy!