Horton

Thread

Commits

  1. Fix handling of empty arrays in array_fill().

  1. BUG #14487: malformed empty array from array_fill

    Andrew Gierth <andrew@tao11.riddles.org.uk> — 2017-01-05T15:21:56Z

    The following bug has been logged on the website:

Bug reference:      14487
Logged by:          Andrew Gierth
Email address:      andrew@tao11.riddles.org.uk
PostgreSQL version: 9.6.1
Operating system:   any
Description:        

array_fill returns a malformed empty array when given a size of zero:

select a, a = '{}' from (select array_fill(null::integer, array[0]) as a)
s;
 a  | ?column? 
----+----------
 {} | f

The result actually has ndims=1 and a length of 0, which can be checked with
array_length(a,1).

Obvious fix is to move the fast-track empty result to after computing
nitems?

  2. Re: BUG #14487: malformed empty array from array_fill

    Andrew Gierth <andrew@tao11.riddles.org.uk> — 2017-01-05T15:33:19Z

    >>>>> "andrew" == andrew  <andrew@tao11.riddles.org.uk> writes:

 andrew> array_fill returns a malformed empty array when given a size of
 andrew> zero:

Also, why does array_fill reject the case of an empty array of
dimensions?  It seems obvious that it should return an empty array
result in such cases.

-- 
Andrew (irc:RhodiumToad)

  3. Re: BUG #14487: malformed empty array from array_fill

    Tom Lane <tgl@sss.pgh.pa.us> — 2017-01-05T15:55:13Z

    Andrew Gierth <andrew@tao11.riddles.org.uk> writes:
> "andrew" == andrew  <andrew@tao11.riddles.org.uk> writes:
>  andrew> array_fill returns a malformed empty array when given a size of
>  andrew> zero:

> Also, why does array_fill reject the case of an empty array of
> dimensions?  It seems obvious that it should return an empty array
> result in such cases.

Yeah.  The code seems to have been intending to allow that:

	ndims = ARR_DIMS(dims)[0];

	if (ndims < 0)				/* we do allow zero-dimension arrays */

but it doesn't work right, precisely because that case comes out
as being zero-dimensional not one dimension of length 0.

Ideally I think we would reject zeroes in the lengths array as being an
error, but it's probably too late for that, because likely people are
using array[0] to get an empty result array since '{}' doesn't work.

It also strikes me that this restriction is quite pointless:

	if (ARR_LBOUND(dims)[0] != 1)
		ereport(ERROR,
				(errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
				 errmsg("wrong range of array subscripts"),
				 errdetail("Lower bound of dimension array must be one.")));

I think it's reasonable to disallow multi-dimensional input arrays here,
but I don't see why it matters what the LB is.

			regards, tom lane