]> granicus.if.org Git - postgresql/commit
Avoid wrong results for power() with NaN input on some platforms.
authorTom Lane <tgl@sss.pgh.pa.us>
Sun, 29 Apr 2018 19:21:45 +0000 (15:21 -0400)
committerTom Lane <tgl@sss.pgh.pa.us>
Sun, 29 Apr 2018 19:21:45 +0000 (15:21 -0400)
commit48e0f8cbe0d9bbc08de6e97ccb6241a670731a97
tree721f6df19308b853c8718e05a82055363a9f9461
parent1d91af81cac75ae6328b0a248c2d478499e4c1d7
Avoid wrong results for power() with NaN input on some platforms.

Per spec, the result of power() should be NaN if either input is NaN.
It appears that on some versions of Windows, the libc function does
return NaN, but it also sets errno = EDOM, confusing our code that
attempts to work around shortcomings of other platforms.  Hence, add
guard tests to avoid substituting a wrong result for the right one.

It's been like this for a long time (and the odd behavior only appears
in older MSVC releases, too) so back-patch to all supported branches.

Dang Minh Huong, reviewed by David Rowley

Discussion: https://postgr.es/m/75DB81BEEA95B445AE6D576A0A5C9E936A73E741@BPXM05GP.gisp.nec.co.jp
src/backend/utils/adt/float.c
src/test/regress/expected/float8-exp-three-digits-win32.out
src/test/regress/expected/float8-small-is-zero.out
src/test/regress/expected/float8-small-is-zero_1.out
src/test/regress/expected/float8.out
src/test/regress/sql/float8.sql