Re: Thoughts on user-defined types for talk at Postgres conference?

Stephen Frost <sfrost(at)snowman(dot)net> writes:
> * Guyren Howe (guyren(at)gmail(dot)com) wrote:
>> It seems as though a Composite Type having only fixed-length fields should be able to be regarded as a fixed-length value.

> While I can understand where you're coming from with that, it's
> certainly not something the system is built to handle today and, at
> least from my recent review of this area, looks like there would be a
> fair bit of code involved to get to that point.

There are some other fundamental problems here:

* To conclude that a tuple of only fixed-length column types is
itself fixed-length, you need an additional assumption that all
the columns are not-null. This is pretty problematic because
we don't even have a way to enforce such a requirement against
generic composite types; we really only promise to enforce
column constraints against columns of physical tables.

* Even if all the semantic enforcement of that were bulletproof,
we'd still have a big problem with compatibility of the on-disk
representation between current releases and a future release
that thought it could elide some part of the tuple header for
all-not-null stored composite values.

I don't say that these problems are insoluble, but they do
look pretty difficult.

regards, tom lane