While working on some code this evening, I came across an interesting problem that was entirely the result of my own doing and thought that I would share this with others.

The problem manifest itself with the failure of information stored in an array within a hash to be updated - For example, the following example illustrates the basic intention of my code, that is, to manipulate an array within a hash element, in this contrived example, to remove web page links from a user profile if the associated checkbox is ticked on a 'delete' form: 

my %user;
eval {
    my $session_id = $cgi->cookie( -name => 'mycookie', -path => '/' )
+;
    tie %user, 'Apache::Session::DB_File', $session_id, { 'FileName' =
+> '.sessions' };
};
croak( 'Cannot retrieve user session ID from client-side cookie' ) if 
+$@;

foreach (0..$#{$user{'links'}}) {
    if ($cgi->param('link_'.$_)) {
        splice(
            @{$user{'links'}},
            $_,
            1
        )
    }
}

[download]

As the code above demonstrates, there is little in the code itself to raise any red flags, however, it simply wasn't working in the larger script - The array within the hash element remained unmodified with execution, when all logic and passed parameters suggested that array elements should be spliced from the data structure.

I freely admit, I was somewhat puzzled by this and pondered on it for some time. Then, as I stared vacantly at my computer screen, I remembered something that I read about tied-hashes long ago in the MLDBM man page - You cannot directly modify an existing reference within a tied hash interface; it must first be retrieved and stored in a temporary variable for further modifications. This limitation, according to the MLDBM man page, is because the Perl TIEHASH interface currently has no support for multi-dimensional ties.

Therein lies the problem with my code - One cannot directly edit data structures stored within a tied-hash. Sure enough, once I updated this section of code to copy the hash element to a temporary array prior to splicing modification, the code worked perfectly. As such, this node, while not illustrative enough on its own to form a tutorial (or necessarily pensive and/or thought-provoking enough to warrant a true meditation), may serve as a solution for others who come across a similar problem.

The lesson learnt - Context is everything when it comes to debugging! :-)

 

Replies are listed 'Best First'.
Re: The importance of context in debugging
by vladb (Vicar) on Jul 26, 2002 at 18:49 UTC
    Indeed so! However, in your particular case it is not necessarily the context but thorough understanding of the tools used that is lacking ;-).

    Same occurred to me just recently when I was playing with the Class::DBI module. My database contained a dozen tables. A few of these tables served as a cross-reference: they linked data from two tables together. So, when I set out to create a Class::DBI for a joint table like so: 
    package YV::Table::ArticleAuthor;
use base 'YV::Table::DBI';

__PACKAGE__->table('article_authors');
__PACKAGE__->columns(All => qw/author_id article_id/);
__PACKAGE__->hasa(YV::Table::Author => 'author_id');
__PACKAGE__->hasa(YV::Table::Article => 'article_id');

# other methods

    [download]
    And later in my code tried to do something like this: 
    use YV::Table::ArticleAuthor;

# this is built dynamically...
my %delete_article_xref = (
                            23 => 1,
                           );

# $author is an YV::Table::Author object initialized prior..
my @author_articles = YV::Table::ArticleAuthor->search(author_id => $a
+uthor->id);

# delete cross reference for $author and article id
# (In affect, I want to 'disassociate' article with $article_id from t
+he $author)
for (@author_articles) {
   next unless $delete_article_xref{$_->article_id};
   $_[0]->delete();
}

    [download]
    Guess what happened next from here?

    Answer (select text):
    As a result of invoking the delete() method on an instance of the YV::Table::ArticleAuthor object, I removed all article to author associations. That is, not just one article from a single (selected) author. This caused a little havoc with a web tool I wrote..

    The actual problem rested with my lack of understanding how Class::DBI worked. Looking back at the documentation, I figured that the reason the delete() method was acting so 'strangely' was because the primary column of the YV::Table::ArticleAuthor class was in fact author_id and delete() uses this very column to construct (or rather 'restrict') it's DELETE SQL. So the quick fix to my problem was to swap the fields like so: 
    __PACKAGE__->columns(All => qw/article_id author_id/);

    [download]
    And since the first column is (by default) taken to be the primary column, delete worked as I wanted it to thereafter. ;-).


    Update: err.. I submitted invalid code initially (was taking it from my head heh). This is fixed now.

    _____________________ 
    # Under Construction

    [download]
[reply]
[d/l]
[select]

Back to Meditations