my $all = $dbh->selectall_arrayref("
    SELECT ID, col1, col2
      FROM table
     WHERE col1 = 'xxx'
");

print "Number of rows selected: ", 0+@$all, "\n";
[download]

my ($count) = $dbh->selectrow_array("
    SELECT COUNT(*)
      FROM table
     WHERE col1 = 'xxx'
");

print "Number of rows selected: $count\n";
[download]

$count = $dbh->selectrow_array('SELECT count(ID) FROM table WHERE col1
+ = ?', \%attr, 'xxx');
print "Number of rows selected: $count ";
[download]

print "Number of rows selected: ", scalar @{$all};
[download]

You have a second problem: You don't use placeholders. That usually means that you are vulnerable to SQL injection, and you prevent any caching of prepared statements.

Always use placeholders, even if the underlying database does not support placeholders. Placeholders are part of the DBI specification, so you can use them with ANY database and pseudo-database supported by DBI. Each database driver (DBD::Whatever) must support placeholders, and together with the DBI, it will take care of proper quoting if needed. Since most database support parameters, there is no need for quoting or modification of the SQL statement, the values are simply passed through a different part of the communication channel to the database, and so the database can optimize and cache much better than with inline values.

I would prefer that DBI issued a warning or even refuse to work with inline values, but that would add a lot of overhead, as DBI would need to actually parse the SQL statement.

Alexander

I would prefer that DBI issued a warning or even refuse to work with inline values, but that would add a lot of overhead, as DBI would need to actually parse the SQL statement.

Worse than that, it would simply be impossible to do.

On receiving the query "SELECT * FROM foo WHERE bar = 'baz'", how would DBI know whether it had been called as $dbh->selectall_arrayref("SELECT * FROM foo WHERE bar = 'baz'"); (which is fine - the baz is a hard-coded literal) or as $dbh->selectall_arrayref("SELECT * FROM foo WHERE bar = '$myvar'"); (which is potentially dangerous)? Even if it could make that determination, in the latter case, how would it know whether $myvar's value came from user input (unsafe) or the statement my $myvar = 'baz'; (another hard-coded literal, so safe)?

We've already got taint mode and you can set DBI to reject tainted values (DBI->connect(..., { TaintIn => 1 })), but that's about as close as you're likely to be able to get.

Note that the SQL statement would be tainted as soon as you interpolate tainted variables into it.

Tainting would also have been my answer to the problem of mixing untrusted user input with SQL fragments.

For constants (your "hard-coded literals"), I really don't care. I've used DBI nearly daily in the last few years, and I rarely needed constants. And for the few cases where I really had constants, I replaced them with placeholders and a constant bind value just because I have the habit to pass all values using placeholders. So having DBI warn or die when it finds a string or number literal in an SQL statement would not hurt much, it would just force you to pass even constants via placeholders. Of course, this "paranoia mode" would break some existing code, and it needs to be disabled by default for a long time. Any by the way: Because no user code needs the $dbh->quote() method in "paranoia mode", DBI should warn or die when non-DBD code calls this method.

I see a different problem that really blocks the implementation of this "paranoia mode": The different SQL dialects. For example, it is possible and sometimes needed to create PL/SQL code via DBI/DBD::Oracle, and that code may legally contain any hard-coded literal. So the actual parsing has to be done in the DBD code, at least for those special cases. DBI code may provide a default parser, for DBDs that don't have those special cases.

Alexander

--
Today I will gladly share my knowledge and experience, for there are no sweeter words than "I told you so". ;-)