randgen team mailing list archive

[Philip Stoev <pstoev@xxxxxxxxx>]

> While it is a powerful and useful bug hunting tool, it seems that we are
> approaching territory where we have useful Transforms that don't apply 
> well
> to every grammar.

It is more about which Transform applies to which generated query and not to 
which grammar.

In the past, failures in Transforms were suppressed, however this caused a 
situation where some 100% non-working Transforms were not detected, 
providing a false sense that stuff is being tested. So, now the balance is 
in the other direction, where most SQL errors in Transforms would cause the 
test to fail with STATUS_ENVIRONMENT_FAILURE.

> For example, the ExecuteAsView Transform does not like subqueries and thus
> breaks with optimizer_subquery.yy.

I just pushed a fix in this where the Transform is not applied if the query 
has several SELECT keywords in its body, which is sort of a poor regexp 
check for subqueries, in the absence of full parsing. Note that each 
Transform can reject any query it does not like outright without processing 
it.

> It is possible to manually alter files and select what Transforms will be
> used (
> http://forge.mysql.com/wiki/RandomQueryGeneratorTransforms#Selecting_specific_Transforms),
> but I was wondering if there was any support for allowing options to be
> passed, for example, an ignore-transform switch or something that would
> eliminate certain ones or even a use-transforms to allow a user to specify 
> a
> single transform (for example if we wanted to only beat up code with
> ExecuteAsSPTwice or something).
>
> In any case, the manual method works, but I just wanted to see what the 
> rest
> of you guys were thinking.

Yes, we are in need of a system where we can pass options to Validators, 
Reporters, Transforms and the like. It has to be integrated with the 
existing system of perl-based configuration files and gnu-based command-line 
options. Any suggestion for a general syntax that we can apply to this and 
future situations would be much appreciated.

Even if the ExecuteAsView situation can be fixed, we will have more of the 
same in the future, as the various database products diverge in the syntax 
and features they support. At the same time, I intend to continue to expand 
the list of Transforms, because I think they are very useful because they 
silently check a lot of stuff people would not normally put in their 
grammars.

Philip Stoev

> Thanks,
> Patrick


--------------------------------------------------------------------------------


> _______________________________________________
> Mailing list: https://launchpad.net/~randgen
> Post to     : randgen@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~randgen
> More help   : https://help.launchpad.net/ListHelp