Re: SQL::Statement confusing literals and identifiers

[grinder]

in reply to SQL::Statement confusing literals and identifiers

SQL::Statement doesn't seem to recognize that "set" means set.

String constants use single quote (') delimiters. Column labels use double quotes ("), and set in your example is a string constant, not a column label.

So the (non-) problem is that it doesn't like parsing illegal SQL. Rewrite your statement as the following, and you should be fine.

SELECT id,
       gid,
       card,
       'set',
       illus,
       num
  FROM Print

[Thilosophy]

No, I think ikegami wants the (double-quoted) column set (that is what the column in his CSV file is called), not a constant string.

[grinder]

Oh duh, silly me, I should have known ikegami would not make a mistake like that. Still, I think SQL::Statement is playing the game correctly. Witness:

use strict;
use warnings;

use SQL::Statement;

my $stmt = SQL::Statement->new(<<END_SQL, SQL::Parser->new);
SELECT id,
             gid,
             card,
             "set",
             illus,
             num
        FROM Print
END_SQL

print join( "\n\t" => $stmt->command, map{ $_->name } $stmt->columns),
+ "\n";

__PRODUCES__
SELECT
        ID
        GID
        CARD
        "set"
        ILLUS
        NUM

Perhaps Print."set" may be advisable. In that case, the output changes slightly:

SELECT
        ID
        GID
        CARD
        "SET"
        ILLUS
        NUM

[grinder]

Ok, to restore my credibility, here's a workaround:

It's a bug alright. At some point the code lines up the requested columns, and the columns the table has to offer. It chokes on the fact that the q{SET} eq q{"SET"} equality fails to hold. It's because at some point the double quotes need to be stripped away, to recover the bare column name underneath and that is not (no longer?) happening.

What is strange is that there appears to be code in there that does just this, but has been commented out. In any event, the following patch seems to do the trick, but one would have to run the full regression test suite to be sure.

--- Statement.pm        Tue Mar 11 13:13:43 2008
+++ /usr/local/lib/perl5/site_perl/5.8.8/SQL/Statement.pm       Tue Ma
+r 11 13:23:05 2008
@@ -1590,6 +1590,7 @@
                        );
        }
        next unless $col;
+       $col =~ s/\A"([^"]+)"\z/$1/;
 ###new
        if (ref $table eq 'SQL::Statement::Table') {
           $table = $table->name;

[ikegami]

That would surely fix the problem in my case, but it's a very incomplete fix. Simply removing the quotes is too minimalistic. That fix would remove some (maybe all) false negatives, but would introduce false positives. Specifically, the following name column pairs become indistinguishable:

• "foo.bar".baz and foo."bar.baz"
• "foo.bar" and foo.bar
• "*" and *

The simplest solution might be to *add* quotes everywhere internally.

That said, I'll use the previously suggested SELECT * rather than applying this incomplete patch.

[bcarroll]

The problem is that the .csv file referenced has fields with quotes in it. Remove the quotes and the errors should go away. Pay attention to the "chunk" that is referenced in the error, that number should be the line number of the problematic text. I spent a lot of time debugging this same error and once I removed the double-quotes (") from the text in the .csv all errors went away.