On 3/25/15 1:21 AM, Pavel Stehule wrote:
> 2015-03-25 0:17 GMT+01:00 Tom Lane <tgl@sss.pgh.pa.us
> <mailto:tgl@sss.pgh.pa.us>>:
>
> Pavel Stehule <pavel.stehule@gmail.com
> <mailto:pavel.stehule@gmail.com>> writes:
> > updated version with Jim Nasby's doc and rebase against last changes in
> > plpgsql.
>
> I started looking at this patch. ISTM there are some pretty
> questionable
> design decisions in it:
>
> 1. Why create a core GUC to control a behavior that's plpgsql-only?
> I think it'd make more sense for this to be a plgsql custom GUC
> (ie, "plpgsql.enable_asserts" or some such name).
>
>
> This type of assertations can be implemented in any PL language - so I
> prefer global setting. But I have not strong option in this case - this
> is question about granularity - and more ways are valid.
+1
> 2. I find the use of errdetail to report whether the assertion condition
> evaluated to FALSE or to NULL to be pretty useless. (BTW, is
> considering
> NULL to be a failure the right thing? SQL CHECK conditions consider
> NULL
> to be allowed ...)
>
>
> This is a question - I am happy with SQL CHECK for data, but I am not
> sure if same behave is safe for plpgsql (procedural) assert. More
> stricter behave is safer - and some bugs in procedures are based on
> unhandled NULLs in variables. So in this topic I prefer implemented
> behave. It is some like:
+1. I think POLA here is that an assert must be true and only true to be
valid. If someone was unhappy with that they could always coalesce(...,
true).
--
Jim Nasby, Data Architect, Blue Treble Consulting
Data in Trouble? Get it in Treble! http://BlueTreble.com