CQL 3.0 with C* 1.1.0~rc1 Tests Fail

[devdazed]

Tests are failing for delete and select by row using C* 1.1.0~rc1 for details see: http://travis-ci.org/#!/simplereach/helenus/jobs/1139784

[ctavan]

This was introduced by bd676d4.

The query being executed is

SELECT * FROM 'cql_test' WHERE id='foobar'

but the correct query would be

SELECT * FROM cql_test WHERE id='foobar'

Apparently C* 1.1 / CQL3 does not allow quoting of CF-names any more.

[ctavan]

I think we should not try to solve this programmatically and just reserve placeholder-replacement to ColumnNames and ColumnValues. This seems to be a common approach with other databases, that use a query-language, like (again) node-mysql.

I'll provide a fix in a minute.

[ctavan]

What a stupid bug: I was not defining the timezone explicitly in CQL date-strings which caused the tests to work on my local laptop but made them fail in travis...

Now works all well :).

[rmechler-sencha]

This issue seems to affect column names too. I.e., it appears that only column value substitution works now using CQL 3.0. (There doesn't appear to be a test that covers this?) I think it would be nice if placeholders continued to work for column names (and even CF names). Maybe auto quoting could be made optional?

[ctavan]

Technically, as long as you still go with CQL 2, you can still use placeholders for column names and CF names with helenus.

In CQL 3, however, "dynamic" column names as we know them from CQL 2 don't exist any more, hence quoting these names will result in CQL parse errors (I'm not 100% an expert on this but that's how I understand the intention of CQL 3. Please correct me if I'm wrong!).

Placeholders are meant to provide an easy and safe way to insert user-provided data into a query. Since there are no more "dynamic" column names in CQL 3 which actually could be considered data in CQL 2, I think it would be wrong to allow placeholders for column names (same goes for CF names).

Also CQL 3 now has prepared CQL statements wich may contain placeholders and which are stored serverside (not yet implemented in helenus, but I'm working on it). These prepared statements don't allow placeholders for CF or column names neither. I think any query containing a placeholder that is accepted by helenus should potentially also work as a prepared statement, so allowing un-quoted placeholders in the case where helenus replaces the placeholders will probably cause a lot of unexpected trouble when trying to run the same queries as prepared statements.

[ctavan]

Ah, you mentioned that there is not test covering the behavior described above: That's right, we leave it to the developer to write proper queries that work with the target version of CQL since assuring the correct usage of placeholders in the queries would mean that we would have to re-implement the entire CQL-parser within helenus which IMO makes not much sense.

[rmechler-sencha]

Thanks for your response.

Ok, fair enough. I'm certainly no expert on CQL 3 myself. It seems to me that even if CQL 3 does not allow dynamic column names, the client side library could support placeholders as a programming convenience (i.e, where the specified column name must be one that exists). Of course the client application could implement something itself to support dynamically generated queries with variable column names if so desired. If you feel it's better not to support placeholders for column names because it goes against the intent of CQL 3, I'm ok with that.

Note, however, that you stated in a previous comment, "... and just reserve placeholder-replacement to ColumnNames and ColumnValues", which did lead me to believe the intent was to support placeholders for column names.

Also note that the CQL example on your main page:

  pool.cql("SELECT '%s' FROM cf_one WHERE key='%s'", ['col','key123'], function(err, results){
    console.log(err, results);
  });

is broken for CQL 3. It might be worth adding a comment to that effect, to save some time for people who are trying it out.

[ctavan]

@rmechler I admit my previous comment was mis-leading. I wrote it before actually fixing the problem and before I found out, that dynamical column names are no longer supported in CQL 3. Please regard that as obsolete.

And yes, the example in the Readme is definitely CQL 2 syntax and won't work with CQL 3 anymore. On the other hand CQL 3 is still beta and will not become the default before Cassandra 1.2, so I'm not sure whether it's already the right time to re-write the documentation. However a comment would be useful, you're right. See #46