Re: Add regression tests for ROLE (USER)

>> I think that it is not that simple: it is a good value to check that the
>> syntax error message conveys a useful information for the user, and that
>> changes to the parser rules do not alter good quality error messages.
>
> It's good to check those things when a feature is implemented.  However,
> once it's done, the odds of the bison parser breaking are very low.

I do not know that. When the next version of bison is out (I have 2.5 from 
2011 on my laptop, 2.7.1 was released on 2013-04-15), or if a new "super 
great acme incredible" drop-in replacement is proposed, you would like to 
see the impact, whether positive or negative, it has on error messages 
before switching.

> Thus, the benefit of testing that over again thousands of times a day
> is pretty tiny.

Sure, I agree that thousands of times per day is an overkill for syntax 
errors. But once in a while would be good, and for that you need to have 
them somewhere, and the current status is "nowhere".

-- 
Fabien.