In dpow(), remove redundant check for whether y is an integer.

I failed to notice that we don't really need to check for y being an
integer in the code path where x = -inf; we already did.

Also make some further cosmetic rearrangements in that spot in hopes
of dodging the seeming compiler bug that buildfarm member fossa is
hitting.  And be consistent about declaring variables as "float8"
not "double", since the pre-existing variables in this function are
like that.

Discussion: https://postgr.es/m/E1jkyFX-0005RR-1Q@gemulon.postgresql.org
This commit is contained in:
Tom Lane 2020-06-16 11:09:42 -04:00
parent 4dd804a99c
commit 400f169373
1 changed files with 10 additions and 11 deletions

View File

@ -1546,7 +1546,7 @@ dpow(PG_FUNCTION_ARGS)
*/
if (isinf(arg2))
{
double absx = fabs(arg1);
float8 absx = fabs(arg1);
if (absx == 1.0)
result = 1.0;
@ -1578,16 +1578,15 @@ dpow(PG_FUNCTION_ARGS)
}
else /* x = -Inf */
{
bool yisoddinteger = false;
/*
* Per POSIX, the sign of the result depends on whether y is an
* odd integer. Since x < 0, we already know from the previous
* domain check that y is an integer. It is odd if y/2 is not
* also an integer.
*/
float8 halfy = arg2 / 2; /* should be computed exactly */
bool yisoddinteger = (floor(halfy) != halfy);
if (arg2 == floor(arg2))
{
/* y is integral; it's odd if y/2 is not integral */
double halfy = arg2 / 2; /* should be computed exactly */
if (halfy != floor(halfy))
yisoddinteger = true;
}
if (arg2 > 0.0)
result = yisoddinteger ? arg1 : -arg1;
else
@ -1622,7 +1621,7 @@ dpow(PG_FUNCTION_ARGS)
result = 0.0; /* we already verified y is positive */
else
{
double absx = fabs(arg1);
float8 absx = fabs(arg1);
if (absx == 1.0)
result = 1.0;