SQL: Signed zero

[philrz]

Postgres happens to support presentation of "signed zero" from the IEEE 754 standard for floating-point arithmetic, e.g.,

$ psql postgres -c "SELECT -0.0::REAL;"
 ?column? 
----------
       -0
(1 row)

Whereas SuperDB currently seems to offer just the one rendering of zero.

$ super -version
Version: 292f4415e

$ super -c "SELECT -0.0::REAL;"
{"-0.0::<float32>":0.::float32}

Details

Repro is with super 292f441. This effect was spotted in sqllogictest random/select/slt_good_50/q3699.

As the standard describes, these values like -0 and +0 are mathematically equivalent to 0, so this effectively seems like just a presentation issue. That said, in addition to causing such sqllogictests to "fail" (albeit in an innocent way that can easily be explained and/or worked around), our docs on Number types point out how SuperSQL supports other concepts from IEEE 754 like Inf, +Inf, -Inf, and NaN, so we might consider supporting -0 and +0 as well for completeness, or document that their absence is intentional.

[philrz]

@mccanne noted that this has implications for the SUP format, since we'd like the format to be able to represent any valid IEEE 754 value.