Tim Bunce wrote:

On Thu, Mar 04, 2004 at 08:56:30AM -0800, Jeff Zucker wrote:


Tim Bunce wrote:


On Thu, Mar 04, 2004 at 01:23:32AM -1000, Beau E. Cox wrote:



On Wednesday 03 March 2004 11:50 pm, Beau E. Cox wrote:




Hello folks,

I noticed changed in DBD::DBM in the latest svn DBI source.

When I try to test DBI (make is fine):





[TEST ERRORS]

Here is what I did to clear the problen on my machine:

1) added dbm to DBI registry (as noted in a prior reply).
2) patched DBM.pm to allow additions of private attribs:

--- trunk/lib/DBD/DBM.pm 2004-03-03 22:03:28.000000000 -1000
+++ test-trunk/lib/DBD/DBM.pm 2004-03-04 01:02:38.000000000 -1000
@@ -113,9 +113,9 @@
#
$var = 'dbm_' . $var unless $var =~ /^dbm_/
or $var eq 'f_dir';
- if (!$this->{f_valid_attrs}->{$var}) {
- die "Invalid attribute in connect: '$var'!\n"
- }
+ #if (!$this->{f_valid_attrs}->{$var}) {
+ # die "Invalid attribute in connect: '$var'!\n"
+ #}
$this->{$var} = $val;
}
}


Here's a better way (driver authors please note):

-                # in the connect string the attr names
-                # can either have dbm_ (or foo_) prepended or not
-                # this will add the prefix if it's missing
-                #
-                $var = 'dbm_' . $var unless $var =~ /^dbm_/
-                                     or     $var eq 'f_dir';
-                if (!$this->{f_valid_attrs}->{$var}) {
-                    die "Invalid attribute in connect: '$var'!\n"
-               }
-               $this->{$var} = $val;

+ # Add dbm_ prefix to attributes that need it
+ # (attributes for DBD::Foo in the Driver DSN portion
+ # of the DSN string are allowed to skip the "foo_" prefix)
+ $var = 'dbm_' . $var if !$valid_attrs->{$var}
+ && $valid_attrs->{"dbm_$var"};
+
+ # place the attribute in $attr (possibly replacing an exiting value)
+ # so DBI->connect will then call our STORE on it for us.
+ # Our STORE method (inherited from DBD::File's) should validate it
+ $attr->{$var} = $val;


That's a better approach I think. I've not checked the test to see
that they're exercising this code though.


Yes, but it needs work in DBD::File. My aim is to make this available to all DBDs that subclass DBD::File without making them subclass STORE and FETCH. Basically if the DBD defines an f_valid_attributes hash, File will do a valid_attributes check on all lower case attributes that start with either foo_ (dbm_ in this case) OR (and this is important) with f_ (for attrs of DBD::File) or with sql_ (for attrs of SQL::Statement or of DBI::SQL::Nano). It will not check other lower-cased attributes since those might have been put in by Proxy or Multiplex, or some other thing. This still seems kind of sloppy to me,


I agree.




there should be some way to make sure that all $dbh->{whatever} stores and fetches use valid keys, otherwise it's too easy to make a typo there and have the module behave unpredictably because you expected something to be set when what you set was something spelled similarly to what you wanted. Ideally all DBDs would check their own attributes for valid key names.



Right, which brings us back to "making them subclass STORE".



But why not just let the DBD::File subclasser define an f_valid_attrs hash and let DBD::File do it like this (only for lower-cased attribute names)


       my $prefix = $dbh->{f_valid_attrs}->{_prefix};
       if ( $prefix
        and !$dbh->{f_valid_attrs}->{$attrib}
        and $attrib =~ /^(f_|sql_|$prefix)/
        ) {
       return $dbh->set_err( 1,"Invalid attribute '$attrib'!");
       }

Has the same effect as what you proposed, but no STORE or FETCH in the subclassed driver.

--
Jeff



Reply via email to