Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitcb3e9e4

Browse files
committed
Put in_range_float4_float8's work in-line.
In commit8b29e88, I'd dithered about whether to makein_range_float4_float8 be a standalone copy of the float in-range logicor have it punt to in_range_float8_float8. I went with the latter, whichsaves code space though at the cost of performance and readability.However, it emerges that this tickles a compiler or hardware bug onbuildfarm member opossum. Test results from commit55e0e45 showconclusively that widening a float4 NaN to float8 produces Inf, not NaN,on that machine; which accounts perfectly for the window RANGE testfailures it's been showing. We can dodge this problem by makingin_range_float4_float8 be an independent function, so that it checksfor NaN inputs before widening them.Ordinarily I'd not be very excited about working around such obviouslybroken functionality; but given that this was a judgment call to beginwith, I don't mind reversing it.
1 parent2f52518 commitcb3e9e4

File tree

1 file changed

+58
-10
lines changed

1 file changed

+58
-10
lines changed

‎src/backend/utils/adt/float.c

Lines changed: 58 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -1254,16 +1254,64 @@ in_range_float8_float8(PG_FUNCTION_ARGS)
12541254
Datum
12551255
in_range_float4_float8(PG_FUNCTION_ARGS)
12561256
{
1257-
/* Doesn't seem worth duplicating code for, so just invoke float8_float8 */
1258-
float8val= (float8)PG_GETARG_FLOAT4(0);
1259-
float8base= (float8)PG_GETARG_FLOAT4(1);
1260-
1261-
returnDirectFunctionCall5(in_range_float8_float8,
1262-
Float8GetDatumFast(val),
1263-
Float8GetDatumFast(base),
1264-
PG_GETARG_DATUM(2),
1265-
PG_GETARG_DATUM(3),
1266-
PG_GETARG_DATUM(4));
1257+
float4val=PG_GETARG_FLOAT4(0);
1258+
float4base=PG_GETARG_FLOAT4(1);
1259+
float8offset=PG_GETARG_FLOAT8(2);
1260+
boolsub=PG_GETARG_BOOL(3);
1261+
boolless=PG_GETARG_BOOL(4);
1262+
float8sum;
1263+
1264+
/*
1265+
* Reject negative or NaN offset. Negative is per spec, and NaN is
1266+
* because appropriate semantics for that seem non-obvious.
1267+
*/
1268+
if (isnan(offset)||offset<0)
1269+
ereport(ERROR,
1270+
(errcode(ERRCODE_INVALID_PRECEDING_FOLLOWING_SIZE),
1271+
errmsg("invalid preceding or following size in window function")));
1272+
1273+
/*
1274+
* Deal with cases where val and/or base is NaN, following the rule that
1275+
* NaN sorts after non-NaN (cf float8_cmp_internal). The offset cannot
1276+
* affect the conclusion.
1277+
*/
1278+
if (isnan(val))
1279+
{
1280+
if (isnan(base))
1281+
PG_RETURN_BOOL(true);/* NAN = NAN */
1282+
else
1283+
PG_RETURN_BOOL(!less);/* NAN > non-NAN */
1284+
}
1285+
elseif (isnan(base))
1286+
{
1287+
PG_RETURN_BOOL(less);/* non-NAN < NAN */
1288+
}
1289+
1290+
/*
1291+
* Deal with infinite offset (necessarily +inf, at this point). We must
1292+
* special-case this because if base happens to be -inf, their sum would
1293+
* be NaN, which is an overflow-ish condition we should avoid.
1294+
*/
1295+
if (isinf(offset))
1296+
{
1297+
PG_RETURN_BOOL(sub ? !less :less);
1298+
}
1299+
1300+
/*
1301+
* Otherwise it should be safe to compute base +/- offset. We trust the
1302+
* FPU to cope if base is +/-inf or the true sum would overflow, and
1303+
* produce a suitably signed infinity, which will compare properly against
1304+
* val whether or not that's infinity.
1305+
*/
1306+
if (sub)
1307+
sum=base-offset;
1308+
else
1309+
sum=base+offset;
1310+
1311+
if (less)
1312+
PG_RETURN_BOOL(val <=sum);
1313+
else
1314+
PG_RETURN_BOOL(val >=sum);
12671315
}
12681316

12691317

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp