Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Peter, should plsh be added to our supplied server-side programming
> languages? Seems like a major feature to me and to others as well.
While plsh is surely a cool hack, I've got considerable unease about
making it into an officially supported feature. The only reasons I can
see for wanting to use it (over plpgsql, pltcl, plperl, etc) are
inherently violations of transaction semantics. Who's going to roll
back your sendmail call when the calling transaction later aborts?
What's going to ensure that the order of external effects has something
to do with the serialization order that the database assigns to several
concurrent transactions?
IMHO plsh is a great tool for shooting yourself in the foot (with a
large-gauge firearm, in fact). People who know what they're doing are
welcome to use it ... but those sorts of people can find and install
it for themselves.
Of course the same worries apply to the untrusted variants of pltcl,
plperl, etc. Perhaps we need to document the risks they carry. But
plsh hasn't even got the possibility of a trusted variant :-(
regards, tom lane
