On 3 April 2013 15:10, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I think though that the upthread argument that we'd have multiple
> interpretations of the same thing is bogus. To me, the core idea that's
> being suggested here is that '{}' should mean a zero-length 1-D array,
> not a zero-D array as formerly. We would still need a way to represent
> zero-D arrays, if only because they'd still exist on-disk in existing
> databases (assuming we're not willing to break pg_upgrade for this).
> I suggest that that ought *not* involve any braces. Perhaps '[]=' would
> be a suitable representation.
Then restoring an old SQL dump would silently convert all old zero-D
arrays to zero-length 1-D arrays. Also, if you failed to completely
rewrite all your application code, and accidentally missed a few
functions, you could very easily end up in a situation where your
tables contained a mixture of zero-D arrays and zero-length 1-D
arrays, that wouldn't compare equally, leading to all sorts of nasty
bugs.
> In the other direction, ISTM that
> '{{},{},{}}' is a zero-by-three array,
Isn't that a three-by-zero array rather than a zero-by-three array?
Either way, what is the use-case for it? I certainly can't remember
ever using a 0x3 or 3x0 matrix.
> entirely distinct from '{}' or
> '{{}}' in dimensionality if not content.
And '{{}}' is a one-by-zero array. There is currently no proposed
syntax for representing a zero-by-zero array, or a zero-by-one array.
All I see here is endless opportunity for confusion and code breakage,
with very little real benefit.
I actually don't think the current behaviour is broken. I find it
perfectly logical and consistent to have a single dimensionless
concept of an empty array. It has no elements, so questions like are
they arranged in 1-D or 2-D, and what index do they start at are
meaningless, and so should return NULL. The only counter-intuitive
thing is that array_length() returns NULL instead of 0 for the empty
array, and it doesn't take long to get used to that behaviour.
Much of this seems to have grown out of a desire to fix the confusing
behaviour of array_length(), but seems to have ended up in a much more
confusing place.
Given that array_length() takes 2 arguments, it is somewhat forgivable
for it to return NULL when you ask for the array's length in a
dimension that it doesn't have. Perhaps what might make the API easier
to work with is a new single-argument function --- e.g., array_size()
that would return the number of elements in the array --- 0 for '{}',
3 for '{1,2,3}', 6 for '{{1,2},{3,4},{5,6}}' and so on.
Regards,
Dean