From: Tom Lane <tgl@sss.pgh.pa.us>
To: ma100@hotmail.com
Cc: pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #17172: NaN compare error in hash agg
Date: Wed, 01 Sep 2021 11:40:36 -0400
Message-ID: <2077133.1630510836@sss.pgh.pa.us> (raw)
In-Reply-To: <17172-7505bea9e04e230f@postgresql.org>
References: <17172-7505bea9e04e230f@postgresql.org>
PG Bug reporting form <noreply@postgresql.org> writes:
> postgres=# select '-NaN'::float union select ('Infinity'::float +
> '-Infinity') union select 'NaN';
> float8
> --------
> NaN
> NaN
> (2 rows)
Hm. The reason for that is that NaN and -NaN are different bitpatterns
(0x7ff8000000000000 vs 0xfff8000000000000) so they produce different
hash values. (On my machine, Inf plus -Inf produces -NaN, which is
not what I'd have expected, but that's what I see.)
Since float8_eq considers all NaNs equal, this is clearly a bug in the
float hash functions. It's simple enough to fix, along the same lines
that we special-case minus zero: check isnan() and return some fixed
value if so. I'm inclined to do that like this:
Datum
hashfloat8(PG_FUNCTION_ARGS)
{
float8 key = PG_GETARG_FLOAT8(0);
/*
* On IEEE-float machines, minus zero and zero have different bit patterns
* but should compare as equal. We must ensure that they have the same
* hash value, which is most reliably done this way:
*/
if (key == (float8) 0)
PG_RETURN_UINT32(0);
+ /*
+ * Similarly, NaNs can have different bit patterns but should compare
+ * as equal. For backwards-compatibility reasons we force them all to
+ * have the hash value of a standard NaN.
+ */
+ if (isnan(key))
+ key = get_float8_nan();
+
return hash_any((unsigned char *) &key, sizeof(key));
}
This has been broken for a very long time. Do we dare back-patch
the fix? Given the lack of complaints, maybe fixing it in HEAD/v14
is enough. OTOH, it's not likely that many people have hash indexes
containing minus NaNs, so maybe it's okay to back-patch.
regards, tom lane
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-bugs@postgresql.org
Cc: tgl@sss.pgh.pa.us, ma100@hotmail.com, pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #17172: NaN compare error in hash agg
In-Reply-To: <2077133.1630510836@sss.pgh.pa.us>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox